All plugins

holographic

Local fact-store memory provider: SQLite with FTS5 keyword search (word forms in English, Russian, Spanish, French, Portuguese, Indonesian, German, Hindi, Arabic, Bengali, Italian, Turkish, Polish and Dutch; Russian and English words of the same root at a lower weight, from bundled DerivBase.Ru (Apache-2.0) and Open English Wordnet (CC BY 4.0) pairs; German compounds and German and Spanish irregular verb forms from bundled Wikidata Lexemes tables (CC0); Chinese, Japanese, Korean and Thai by character n-grams, Chinese also split into words with jieba's dictionary (MIT), Traditional and Simplified Chinese compared as one through a bundled OpenCC table (Apache-2.0), Japanese spelling variants through SudachiDict's (Apache-2.0); synonyms through keywords the agent gives when it stores a fact), trust scores, entities with aliases and HRR (holographic reduced representation) recall, with no API key, network or model calls unless an optional online feature is switched on: search by meaning (embeddings: on) or rescoring by a cross-encoder (rerank: on), both off by default. The provider that shipped in-tree as plugins/memory/holographic, with the same provider name, config (plugins.hermes-memory-store) and memory_store.db, so hermes update moves existing users onto it; a file without merges keeps history schema 1; the first merge raises it to schema 2, after which writers of 0.15-0.18 refuse it until 0.19.0 or a pre-merge backup is restored. Taken over from Nous Research's handoff copy (NousResearch/hermes-plugin-holographic) and maintained by itpartypattaya; since 0.1.1 it fixes probe/reason/related/contradict, non-atomic writes and the never-written retrieval_count, gates recall by relevance without repeats, flags duplicates and conflicts when a fact is added, and honours the provider contract (read-only subagent contexts, hermes_home, mirror metadata, backup_paths); since 0.8.0 it keeps the history of every fact (change / end / correct / revert, search and list as_of a date) and forget erases a fact with all its versions from the file; since 0.16.0 a name guessed from a fact's text must be shaped like one and stays provisional until a second fact or a person confirms it, and entities can be renamed or deleted (the facts stay). Since 0.18.0 an optional, experimental second stage reorders the recall candidates with a cross-encoder at a /v1/rerank endpoint that answers in Cohere's shape without a key, such as llama-server (rerank: on, off by default). It does not raise hits: on the maintainer's held-out set (126 questions) it cut non-target facts in the top five by about 40 % and lines per turn by about a third, and lost about one target per hundred questions. Disclosure — local only unless an optional online feature (search by meaning or rescoring) is switched on: reads and writes one SQLite file per profile ($HERMES_HOME/memory_store.db unless db_path says otherwise), or with scope: user / chat / thread one file per user or chat in holographic-users/ or holographic-chats/ next to it; each fact version records where it was written (platform, chat type, chat and thread id, session id, the turn author's id and display name, and whether the person said it or the agent inferred it; never chat titles), shown in full by list, history and export and not in recall unless recall_source: short (date, display name, this or another chat); each confirmation of a fact by a person keeps its time, session id and that person's id (no text); forget deletes that and the confirmations with the fact and keeps only an HMAC of the text, keyed by a random key stored in the same file, as a tombstone; delete_entity removes a name, its aliases and links from the entity index (never a fact) and keeps an HMAC of each name the same way (not of an alias only a fact's text gave, which goes without one), and both rewrite the file so freed pages keep no copy; the first writable open after an upgrade that changes the extraction rules relinks the facts to their entities under the new rules (stores up to 2,000 facts and 500,000 characters; larger ones on fact_store rebuild_entities), changing only the names it guessed and their links, never a fact, a declared name or an alias someone gave; subagents never write, but a cron job's explicit fact_store / fact_feedback calls do unless cron_writes: false (on by default); no subprocesses or telemetry (at the debug log level the recall stages log fact ids and scores, never text), and no network calls or credentials unless embeddings: on or rerank: on (both off by default). With embeddings: on, the text of facts, and of messages and fact_store search queries that are not greetings, goes to the OpenAI-compatible embeddings endpoint the user configures, also after cron turns while cron_writes is on (embedding_url; https, plain http only to localhost, a private or link-local address, or a local host name such as one without a dot or ending in .local), with the key read from the environment variable the user names (embedding_key_env, OPENAI_API_KEY unless set, sent as a Bearer token to whatever embedding_url is configured, never stored in config) and no redirects followed; facts that recall would not inject, and facts and messages that look like they hold a secret (checked by pattern, best effort), are not sent; the vectors are kept in the same file (table fact_embeddings), deleted with their fact by forget / end / remove and all of them by the next writable session after the option is switched off; the provider keeps what it received under its own terms; with rerank: on, before a turn the user's message and up to rerank_top_k (40) candidate fact texts, and for a fact_store search its query and results, go to the rerank endpoint the user configures (rerank_url; https, plain http only to localhost, a private or link-local address, or a local host name), with no credentials and no redirects followed; facts that recall would not inject or that look like they hold a secret, and queries that hold one, are not sent; a failure or a slow answer gives the recall of rerank: off and pauses the endpoint (60 s, doubling to 900 s); nothing the endpoint returns is stored; the bundled language dictionaries are gzip-compressed plain-text word lists; with auto_extract on (off by default) user sentences that state a preference or decision are stored. Since 0.19.0, fact_store merge combines duplicate facts in one transaction while keeping each copy's history, sources and confirmations under its original id. The retained fact keeps its own text, trust and origin; copies leave current search and recall. Optional expect pins the exact shown versions and refuses changed participants without a write. Scheduled jobs can merge only copies created in their own session. Forgetting the retained fact also erases its merged copies. Merge is local and makes no new network or model calls. Version 0.19.1 adds strict tooling for comparing recorded margins; the normal RECALL_MARGIN remains 1.0. Version 0.19.2 closes a timed-out rerank or embeddings request socket so a server dripping response headers does not keep its worker alive. During TLS wrapping, cancellation is retained but the handshake can continue until its own socket timeout; DNS resolution is outside socket cancellation. Version 0.20.0 adds an opt-in local task-outcome journal and the memory-provider tool fact_experience: emitted and candidate fact/version identifiers, an optional short task and correction, outcome/reason and optional corrected-source references. It is separate from trust, origin and ranking. Agent reports are not authenticated human confirmations. Collection and detailed tracing are off by default; retention is bounded to 30 days and 1000 attempts per store, with up to 10 revisions per attempt. Forgetting any referenced fact removes the dependent attempt and outcome from the store. No new model, network or credential use; synchronous local writes add latency. PR #45 also loads NumPy on first use outside the plugin-loader lock; concurrent first callers wait for import completion. Version 0.20.1 limits SQLite busy waits for the optional journal and retries transient contention once. Failed receipts are reported and are not queued; the SQLite timeout is not a full search deadline. Version 0.20.2 writes no receipt for a prefetch that emits nothing, applies journal retention whenever a writer opens the store (also with the journal off) and scrubs removed outcome text from the file like a forgotten fact. Version 0.20.3 adds one system-prompt line, only with the journal on, asking the agent to record an outcome when the person corrects an answer that used recalled facts (an agent report, not human confirmation). Version 0.21.0 adds an opt-in prefetch_max_tokens budget for the recalled block: facts that do not fit are left out whole, never cut; off by default. Version 0.22.0 keeps entity links honest on a text change: a new entities list replaces the declared names, and a declared name the new text no longer says is dropped; revert restores a version's declared names from an additive version_names table (names with keyed hashes, removed with the version and by delete_entity in every spelling a tombstone covers, including by older releases through triggers). Version 0.22.1 makes the secret/instruction screen about 30× faster on long texts with the same decisions (patterns unchanged; a prefilter chooses where they are tried). Version 0.22.2 sends a slow reranker only as many of the first candidates as its measured speed answers in half rerank_timeout (at least two), so a large pool no longer times out and pauses the endpoint; the candidates not sent keep their order behind the scored ones. Version 0.23.0 stops feedback (helpful/unhelpful) from refreshing a fact's age: optional temporal decay counts from when the current version was recorded, and updated_at changes only when the fact itself does; a correction with a new text (correct / update content) starts at its origin's trust without the old text's helpful marks, recalls and confirmations, like change, while a date-only correction keeps them and revert restores the old trust. Version 0.23.1 takes a correction that journal retention removed off the disk on the open connection (the file rewrite runs after the receipt write, with busy timeout 0 at every step) and clears only the rewrite mark it served. Version 0.24.0 adds two experimental features, both off by default and with no new network, model or credential use: memory_map.py, a command-line exporter nothing in the provider calls, which prints one store as a read-only JSON snapshot for a person to read (opened mode=ro in one read transaction; a forgotten fact is its id and date only; corrected versions and texts the safety screen hides carry no text; with --config it reads config.yaml's search settings, with --hermes-home the hermes-dreaming files about the profile store), whose output holds the facts' texts and so is personal data; and prefetch_clause_coverage: on, which gives each of two separate Russian or English questions in one message a second look at the facts the coverage gate left out. Version 0.25.0 limits the memory map export to 1000 facts and 300 names by default (--max-facts / --max-entities, 0 for none) and says what it cut in SNAP.limits; nothing else changes. Version 0.25.1 fixes recall by meaning dropping a fact that names a short Russian name in another grammatical case than the message (Eva as «Ева» / «Еву»): the name now also meets a capitalized word of the fact that may be a case form of it. Version 0.25.2 reads that word by its case ending rather than a short stem: names whose stem changes meet their forms («Лев» / «Льва»), and different names that only shared a short stem («Аня» / «Анна») no longer meet; nothing else changes. Version 0.25.3 no longer imports NumPy when the provider starts (the first search that needs it does, as does opening a store whose vectors need re-encoding) and treats a NumPy that is installed but fails to import as missing, so recall goes on by words; finds a Latin word written into Chinese, Japanese, Korean or Thai text by itself; refuses at write, and leaves out of recall, a prompt addressed to an assistant and the agent's report of its own steps; and auto_extract (off by default) no longer stores a reply such as "I like that" as a preference; no change to defaults, schema or network behaviour. Version 0.26.0 adds an experimental presence gate, prefetch_presence_gate (off by default): with it on, a fact this provider injected is held back past the six-turn repeat window only while the model can verifiably still see it (its exact recall line inside a <memory-context> block the core fenced, or its fact id and text in a fact_store result), reading only the messages the core already passes to sync_turn, with no new storage, network, model or credential use; tool results now carry raw Unicode instead of Unicode escape sequences (a payload that cannot be encoded keeps the escaped form); change names the keywords it drops and correct writes the keywords it is given; with the gate off, recall is unchanged. Version 0.26.1 fixes that gate after an independent review: a sync the core queued before a compaction no longer counts as the current view, a fact the compaction removed is recalled again once a fresh view no longer shows it, a compaction that fails no longer switches the gate off for the session, a late sync keeps its own turn, and a fact_store history result counts as the fact being visible; with the gate off nothing changes. Version 0.26.2: with the task-outcome journal on (off by default), a prefetch that only refreshes facts the session already holds records no new attempt and carries no Memory attempt line, so a repeated block adds nothing new to the context. Version 0.26.3: with the journal and the presence gate on, a delivery whose attempt could not be written after a compaction no longer stops that fact from recording attempts later. Version 0.26.4 fixes the presence gate after a second independent review: two late syncs in a row each keep their own turn, a quiet session that compacted leaves no lasting state, and a refresh no longer keeps token-budget room for an attempt line it does not carry. Version 0.26.5 fixes four edge cases found in a third review of the presence gate and the journal (both off by default): a long pasted message keeps its turn, a session the provider forgot drops its turns' messages, a failed attempt write drops only receipts older than the last compaction, and the token budget no longer keeps room for an attempt line in a block that became a refresh. Version 0.26.6 loads numpy's OpenBLAS on one thread on Windows when the plugin is what imports numpy, which saves about 225 MB of commit charge per Hermes process on an 8-thread machine. Version 0.27.0 adds French auto-extraction of stated preferences and shared decisions, and an optional one-hop memory graph over the entity index that reaches facts next to the ones a query names (bounded reads, off by default). Version 0.28.0 links a wiki pointer line (Wiki page: namespace/.../slug.md) to its namespace and page slug as entities, and recall before a turn searches only what the person wrote: a reply's quoted message, attachment notes, the author's name prefix in shared chats and the cron hint are left out, and a notice of finished background work recalls nothing. Version 0.29.0 records in the optional task-outcome journal (with experience_trace on) one word for why the reranker did or did not order a recall, never error text or facts, and adds measurement tooling; no change to defaults, schema or network behaviour. Version 0.29.1 refuses a fact id that is not a whole number (a JSON true or 1.9 acted on fact #1, and forget erased it) and restores the connection's own SQLite busy timeout after the recall-count flush; no change to defaults, schema or network behaviour. Version 0.29.2 refuses a trust change, trust floor or limit that is no number (a JSON true or "nan" in trust_delta set a fact's trust to the top, and min_trust: true made search and list answer nothing); no change to defaults, schema or network behaviour.

itpartypattaya42Memoryv0.29.2Updated 13 hours ago

About

Local fact-store memory provider: SQLite with FTS5 keyword search (word forms in English, Russian, Spanish, French, Portuguese, Indonesian, German, Hindi, Arabic, Bengali, Italian, Turkish, Polish and Dutch; Russian and English words of the same root at a lower weight, from bundled DerivBase.Ru (Apache-2.0) and Open English Wordnet (CC BY 4.0) pairs; German compounds and German and Spanish irregular verb forms from bundled Wikidata Lexemes tables (CC0); Chinese, Japanese, Korean and Thai by character n-grams, Chinese also split into words with jieba's dictionary (MIT), Traditional and Simplified Chinese compared as one through a bundled OpenCC table (Apache-2.0), Japanese spelling variants through SudachiDict's (Apache-2.0); synonyms through keywords the agent gives when it stores a fact), trust scores, entities with aliases and HRR (holographic reduced representation) recall, with no API key, network or model calls unless an optional online feature is switched on: search by meaning (embeddings: on) or rescoring by a cross-encoder (rerank: on), both off by default. The provider that shipped in-tree as plugins/memory/holographic, with the same provider name, config (plugins.hermes-memory-store) and memory_store.db, so hermes update moves existing users onto it; a file without merges keeps history schema 1; the first merge raises it to schema 2, after which writers of 0.15-0.18 refuse it until 0.19.0 or a pre-merge backup is restored. Taken over from Nous Research's handoff copy (NousResearch/hermes-plugin-holographic) and maintained by itpartypattaya; since 0.1.1 it fixes probe/reason/related/contradict, non-atomic writes and the never-written retrieval_count, gates recall by relevance without repeats, flags duplicates and conflicts when a fact is added, and honours the provider contract (read-only subagent contexts, hermes_home, mirror metadata, backup_paths); since 0.8.0 it keeps the history of every fact (change / end / correct / revert, search and list as_of a date) and forget erases a fact with all its versions from the file; since 0.16.0 a name guessed from a fact's text must be shaped like one and stays provisional until a second fact or a person confirms it, and entities can be renamed or deleted (the facts stay). Since 0.18.0 an optional, experimental second stage reorders the recall candidates with a cross-encoder at a /v1/rerank endpoint that answers in Cohere's shape without a key, such as llama-server (rerank: on, off by default). It does not raise hits: on the maintainer's held-out set (126 questions) it cut non-target facts in the top five by about 40 % and lines per turn by about a third, and lost about one target per hundred questions. Disclosure — local only unless an optional online feature (search by meaning or rescoring) is switched on: reads and writes one SQLite file per profile ($HERMES_HOME/memory_store.db unless db_path says otherwise), or with scope: user / chat / thread one file per user or chat in holographic-users/ or holographic-chats/ next to it; each fact version records where it was written (platform, chat type, chat and thread id, session id, the turn author's id and display name, and whether the person said it or the agent inferred it; never chat titles), shown in full by list, history and export and not in recall unless recall_source: short (date, display name, this or another chat); each confirmation of a fact by a person keeps its time, session id and that person's id (no text); forget deletes that and the confirmations with the fact and keeps only an HMAC of the text, keyed by a random key stored in the same file, as a tombstone; delete_entity removes a name, its aliases and links from the entity index (never a fact) and keeps an HMAC of each name the same way (not of an alias only a fact's text gave, which goes without one), and both rewrite the file so freed pages keep no copy; the first writable open after an upgrade that changes the extraction rules relinks the facts to their entities under the new rules (stores up to 2,000 facts and 500,000 characters; larger ones on fact_store rebuild_entities), changing only the names it guessed and their links, never a fact, a declared name or an alias someone gave; subagents never write, but a cron job's explicit fact_store / fact_feedback calls do unless cron_writes: false (on by default); no subprocesses or telemetry (at the debug log level the recall stages log fact ids and scores, never text), and no network calls or credentials unless embeddings: on or rerank: on (both off by default). With embeddings: on, the text of facts, and of messages and fact_store search queries that are not greetings, goes to the OpenAI-compatible embeddings endpoint the user configures, also after cron turns while cron_writes is on (embedding_url; https, plain http only to localhost, a private or link-local address, or a local host name such as one without a dot or ending in .local), with the key read from the environment variable the user names (embedding_key_env, OPENAI_API_KEY unless set, sent as a Bearer token to whatever embedding_url is configured, never stored in config) and no redirects followed; facts that recall would not inject, and facts and messages that look like they hold a secret (checked by pattern, best effort), are not sent; the vectors are kept in the same file (table fact_embeddings), deleted with their fact by forget / end / remove and all of them by the next writable session after the option is switched off; the provider keeps what it received under its own terms; with rerank: on, before a turn the user's message and up to rerank_top_k (40) candidate fact texts, and for a fact_store search its query and results, go to the rerank endpoint the user configures (rerank_url; https, plain http only to localhost, a private or link-local address, or a local host name), with no credentials and no redirects followed; facts that recall would not inject or that look like they hold a secret, and queries that hold one, are not sent; a failure or a slow answer gives the recall of rerank: off and pauses the endpoint (60 s, doubling to 900 s); nothing the endpoint returns is stored; the bundled language dictionaries are gzip-compressed plain-text word lists; with auto_extract on (off by default) user sentences that state a preference or decision are stored. Since 0.19.0, fact_store merge combines duplicate facts in one transaction while keeping each copy's history, sources and confirmations under its original id. The retained fact keeps its own text, trust and origin; copies leave current search and recall. Optional expect pins the exact shown versions and refuses changed participants without a write. Scheduled jobs can merge only copies created in their own session. Forgetting the retained fact also erases its merged copies. Merge is local and makes no new network or model calls. Version 0.19.1 adds strict tooling for comparing recorded margins; the normal RECALL_MARGIN remains 1.0. Version 0.19.2 closes a timed-out rerank or embeddings request socket so a server dripping response headers does not keep its worker alive. During TLS wrapping, cancellation is retained but the handshake can continue until its own socket timeout; DNS resolution is outside socket cancellation. Version 0.20.0 adds an opt-in local task-outcome journal and the memory-provider tool fact_experience: emitted and candidate fact/version identifiers, an optional short task and correction, outcome/reason and optional corrected-source references. It is separate from trust, origin and ranking. Agent reports are not authenticated human confirmations. Collection and detailed tracing are off by default; retention is bounded to 30 days and 1000 attempts per store, with up to 10 revisions per attempt. Forgetting any referenced fact removes the dependent attempt and outcome from the store. No new model, network or credential use; synchronous local writes add latency. PR #45 also loads NumPy on first use outside the plugin-loader lock; concurrent first callers wait for import completion. Version 0.20.1 limits SQLite busy waits for the optional journal and retries transient contention once. Failed receipts are reported and are not queued; the SQLite timeout is not a full search deadline. Version 0.20.2 writes no receipt for a prefetch that emits nothing, applies journal retention whenever a writer opens the store (also with the journal off) and scrubs removed outcome text from the file like a forgotten fact. Version 0.20.3 adds one system-prompt line, only with the journal on, asking the agent to record an outcome when the person corrects an answer that used recalled facts (an agent report, not human confirmation). Version 0.21.0 adds an opt-in prefetch_max_tokens budget for the recalled block: facts that do not fit are left out whole, never cut; off by default. Version 0.22.0 keeps entity links honest on a text change: a new entities list replaces the declared names, and a declared name the new text no longer says is dropped; revert restores a version's declared names from an additive version_names table (names with keyed hashes, removed with the version and by delete_entity in every spelling a tombstone covers, including by older releases through triggers). Version 0.22.1 makes the secret/instruction screen about 30× faster on long texts with the same decisions (patterns unchanged; a prefilter chooses where they are tried). Version 0.22.2 sends a slow reranker only as many of the first candidates as its measured speed answers in half rerank_timeout (at least two), so a large pool no longer times out and pauses the endpoint; the candidates not sent keep their order behind the scored ones. Version 0.23.0 stops feedback (helpful/unhelpful) from refreshing a fact's age: optional temporal decay counts from when the current version was recorded, and updated_at changes only when the fact itself does; a correction with a new text (correct / update content) starts at its origin's trust without the old text's helpful marks, recalls and confirmations, like change, while a date-only correction keeps them and revert restores the old trust. Version 0.23.1 takes a correction that journal retention removed off the disk on the open connection (the file rewrite runs after the receipt write, with busy timeout 0 at every step) and clears only the rewrite mark it served. Version 0.24.0 adds two experimental features, both off by default and with no new network, model or credential use: memory_map.py, a command-line exporter nothing in the provider calls, which prints one store as a read-only JSON snapshot for a person to read (opened mode=ro in one read transaction; a forgotten fact is its id and date only; corrected versions and texts the safety screen hides carry no text; with --config it reads config.yaml's search settings, with --hermes-home the hermes-dreaming files about the profile store), whose output holds the facts' texts and so is personal data; and prefetch_clause_coverage: on, which gives each of two separate Russian or English questions in one message a second look at the facts the coverage gate left out. Version 0.25.0 limits the memory map export to 1000 facts and 300 names by default (--max-facts / --max-entities, 0 for none) and says what it cut in SNAP.limits; nothing else changes. Version 0.25.1 fixes recall by meaning dropping a fact that names a short Russian name in another grammatical case than the message (Eva as «Ева» / «Еву»): the name now also meets a capitalized word of the fact that may be a case form of it. Version 0.25.2 reads that word by its case ending rather than a short stem: names whose stem changes meet their forms («Лев» / «Льва»), and different names that only shared a short stem («Аня» / «Анна») no longer meet; nothing else changes. Version 0.25.3 no longer imports NumPy when the provider starts (the first search that needs it does, as does opening a store whose vectors need re-encoding) and treats a NumPy that is installed but fails to import as missing, so recall goes on by words; finds a Latin word written into Chinese, Japanese, Korean or Thai text by itself; refuses at write, and leaves out of recall, a prompt addressed to an assistant and the agent's report of its own steps; and auto_extract (off by default) no longer stores a reply such as "I like that" as a preference; no change to defaults, schema or network behaviour. Version 0.26.0 adds an experimental presence gate, prefetch_presence_gate (off by default): with it on, a fact this provider injected is held back past the six-turn repeat window only while the model can verifiably still see it (its exact recall line inside a <memory-context> block the core fenced, or its fact id and text in a fact_store result), reading only the messages the core already passes to sync_turn, with no new storage, network, model or credential use; tool results now carry raw Unicode instead of Unicode escape sequences (a payload that cannot be encoded keeps the escaped form); change names the keywords it drops and correct writes the keywords it is given; with the gate off, recall is unchanged. Version 0.26.1 fixes that gate after an independent review: a sync the core queued before a compaction no longer counts as the current view, a fact the compaction removed is recalled again once a fresh view no longer shows it, a compaction that fails no longer switches the gate off for the session, a late sync keeps its own turn, and a fact_store history result counts as the fact being visible; with the gate off nothing changes. Version 0.26.2: with the task-outcome journal on (off by default), a prefetch that only refreshes facts the session already holds records no new attempt and carries no Memory attempt line, so a repeated block adds nothing new to the context. Version 0.26.3: with the journal and the presence gate on, a delivery whose attempt could not be written after a compaction no longer stops that fact from recording attempts later. Version 0.26.4 fixes the presence gate after a second independent review: two late syncs in a row each keep their own turn, a quiet session that compacted leaves no lasting state, and a refresh no longer keeps token-budget room for an attempt line it does not carry. Version 0.26.5 fixes four edge cases found in a third review of the presence gate and the journal (both off by default): a long pasted message keeps its turn, a session the provider forgot drops its turns' messages, a failed attempt write drops only receipts older than the last compaction, and the token budget no longer keeps room for an attempt line in a block that became a refresh. Version 0.26.6 loads numpy's OpenBLAS on one thread on Windows when the plugin is what imports numpy, which saves about 225 MB of commit charge per Hermes process on an 8-thread machine. Version 0.27.0 adds French auto-extraction of stated preferences and shared decisions, and an optional one-hop memory graph over the entity index that reaches facts next to the ones a query names (bounded reads, off by default). Version 0.28.0 links a wiki pointer line (Wiki page: namespace/.../slug.md) to its namespace and page slug as entities, and recall before a turn searches only what the person wrote: a reply's quoted message, attachment notes, the author's name prefix in shared chats and the cron hint are left out, and a notice of finished background work recalls nothing. Version 0.29.0 records in the optional task-outcome journal (with experience_trace on) one word for why the reranker did or did not order a recall, never error text or facts, and adds measurement tooling; no change to defaults, schema or network behaviour. Version 0.29.1 refuses a fact id that is not a whole number (a JSON true or 1.9 acted on fact #1, and forget erased it) and restores the connection's own SQLite busy timeout after the recall-count flush; no change to defaults, schema or network behaviour. Version 0.29.2 refuses a trust change, trust floor or limit that is no number (a JSON true or "nan" in trust_delta set a fact's trust to the top, and min_trust: true made search and list answer nothing); no change to defaults, schema or network behaviour.

Holographic Memory Provider

Local memory provider for Hermes Agent: a SQLite fact store with
FTS5 search, trust scoring, entity resolution and HRR (holographic reduced representation) retrieval. By default
everything stays on your machine: no API key, no network, no model calls. Search by
meaning
is an optional online add-on, off unless you turn it on.

Maintained by itpartypattaya since October 2026. This is the provider that
shipped inside Hermes core as plugins/memory/holographic/; Nous Research
handed it off when the bundled
copy was scheduled to leave core (announced for October 15, 2026). Same provider name, same config, same database:
users who have memory.provider: holographic keep their facts. See CHANGELOG.md for what changed.

Task outcomes and memory explanations (0.20.0)

An accurate fact can still be the wrong answer to a task — for example, a fact about another person with the same
name. The optional local experience journal records the actual fact versions returned, the task outcome and a
short correction separately from truth/trust. It does not change ranking or automatically correct a fact.

Add to plugins.hermes-memory-store in config.yaml, then restart the gateway or start a new CLI session:

experience: on
experience_trace: on  # optional: candidate/stage ids as well as the actual emission receipt

Both are off by default. When enabled, fact_experience provides recent, outcome and explain; a search returns
memory_attempt, and prefetch identifies a nonempty emission; a prefetch that recalls nothing writes nothing.
With collection on, one line in the system prompt asks the agent to record an outcome when the person corrects or rejects
an answer that used recalled facts (0.20.3); with it off the prompt is unchanged.
Explain reads that original snapshot without replaying search or calling a model. Task outcomes are agent reports, never authenticated human confirmations; old
fact_feedback calls keep their original meaning and are not imported as duplicate evidence.

Records expire after 30 days, with at most 1000 attempts per scoped database and 10 revisions per attempt; the limits
also apply when a writer opens the store with collection turned off. Forgetting
a referenced fact/version also erases the dependent task/correction; merge and end preserve history. Detailed tracing
costs local storage and synchronous writes; no fixed latency or token-saving percentage is promised.
If the journal encounters a busy SQLite writer, it retries briefly with limited SQLite busy waits. After repeated
failure the search still returns, but that attempt has no memory_attempt and cannot
receive an outcome. Slow-operation warnings contain stage timings only, not fact or query text. This is best-effort
recording, not a guarantee against long locks, disk failures or process termination. See the full behavior.
Dreaming 3.6.0 can read live summaries without another correction-text cache. See the full contract and examples.

How it works

Architecture: before each turn Hermes core sends the message through the relevance gate and search; writes pass a
validator that refuses instructions and credentials and land in one SQLite transaction with the fact's history; recall
and history tools read the same file; search by meaning is an optional online add-on

Recall before a turn: greetings recall nothing; the message's words become keys; 15 candidates come from FTS5,
character n-grams or a word scan, reordered by meaning if that is on; a relevance gate, a repeat window and the screen
decide; up to five facts by their words, one by meaning alone and two of named entities go into the turn as
data

The graph (graph_expansion, off by default) is one bounded hop: a fact that names two entities makes those entities
neighbours, and a search walks a single hop out of the entities its words resolve to, so a fact about a neighbouring
entity is reachable without sharing a word with the query. Hub entities — a namespace, a project every fact mentions —
are never traversed, and inside search() a discovery only fills the tail of the result list, never displacing a text
hit (a reranker run afterwards may reorder the pool). Bounds, measurements and repair: docs/memory-graph.md.

Install

While your Hermes still bundles holographic, the bundled copy is used and this plugin stays dormant. After core
drops it, hermes update installs the catalog plugin named holographic for every profile that has it configured.
To install by hand:

hermes plugins install holographic          # from the plugin catalog
hermes config set memory.provider holographic

or hermes memory setup and pick holographic.

The catalog entry may still pin an older release while a version bump waits for review; to install a given release
exactly, pass its commit: hermes plugins install holographic --ref <40-character sha of the tag> --force.

Memory went off after hermes update

Core dropped its bundled copy on 2026-10-11 (Hermes v0.21.7) and installs this plugin in its place. If that
install fails (a dependency resolution error in the plugin workspace, a network error), Hermes reports "Memory
provider 'holographic' is not installed, so external memory is off", and a failed install may leave memory.provider
empty and holographic listed under plugins.disabled in config.yaml. Your facts are not touched:
memory_store.db stays where it was. To bring memory back:

  1. hermes plugins install holographic (add --force if a copy is already in ~/.hermes/plugins/holographic/);
    if it fails on dependencies, update Hermes first — the cause is usually in core's lock, not in this plugin.
  2. In config.yaml: remove holographic from plugins.disabled, and set memory.provider: holographic.
  3. Restart the gateway (or start a new CLI session) and check with hermes memory status.

Config

In config.yaml under plugins.hermes-memory-store:

KeyDefaultDescription
db_path$HERMES_HOME/memory_store.dbSQLite database path
auto_extractfalseStore user sentences that state a preference or decision (English, Russian, Chinese, French) after each completed turn, and also before context compression and at session end (not in scope: user, where a whole transcript cannot be attributed), at trust 0.4; questions, replies to the assistant ("I like that, sounds good") and restatements of a stored fact are skipped
mirror_memory_writesfalseAlso store every built-in memory add (MEMORY.md / USER.md) as a fact
default_trust0.5Trust score of a new fact
min_trust_threshold0.3Facts below this trust are not recalled
temporal_decay_half_life0Days; 0 turns recency decay off. Age counts from when the fact's current version was recorded: feedback, keywords and tags do not make a fact younger
hrr_dim1024HRR vector dimensions
hrr_weight0.3Weight of the HRR term in search ranking
prefetch_min_coverage0.3Share of a message's informative words (rarer words weigh more) a fact must match to be recalled before the turn; a fact sharing one word of a longer message is recalled only when no other fact holds that word or it may be a name (capitalized inside the sentence, or in a script without capitals)
prefetch_clause_coverageoffExperimental. on: a message that asks two separate questions (Russian or English) gives each of them a second look at the facts the coverage gate left out, so the answer to the shorter question is not lost. Only for a fact that holds two words of its question and names the person that question is about. Not in the setup wizard; a value it does not recognise reads as off. See Two questions in one message
graph_expansionoffon: search adds facts one co-occurrence edge out of the entities the message names, as discoveries in the tail of the result list. Off by default until it is measured on a real store; not in the setup wizard, and a value it does not recognise reads as off. See docs/memory-graph.md
prefetch_repeat_turns6A fact recalled before a turn is not injected again for this many turns of the session
prefetch_max_tokens0Approximate tokens the block recalled before a turn may take, header and every line included (0.21.0). The budget never shortens a line: a fact whose line does not fit is left out whole; the rest keep their order. The older 300-character cap on each prefetch line still applies with or without a budget, so a longer fact's line ends in "…"; stats counts what was left out. 0 = off: up to five facts as before. See docs/context-budget.md
recall_sourceoffshort: each fact recalled before a turn also says where it was written (recorded 2026-10-12 · Anton · this chat; no ids). Off by default: it costs tokens and tells the room which chat a fact came from. list, history and export always show the source in full
cron_writestrueA cron job's explicit fact_store / fact_feedback calls may write (nightly memory jobs such as hermes-dreaming's extraction and keywords exist to make them). Automatic writes (auto-extraction, mirroring, recall counts) stay off in cron, subagents and flush contexts; with embeddings on, a cron job that may write also makes missing fact vectors; false makes cron read-only too
recall_indicatortrueReport recalled facts to the core's recall indicator ("🧠 Holographic — recalled 2 memories"). On Telegram and other messaging platforms that line arrives as a separate message before the reply; false turns it off, recall itself is unchanged. Like every key here it is read when the gateway starts: a change takes effect after a restart
embeddingsoffon: search by meaning through an embeddings API you name (embedding_url, embedding_model, embedding_key_env, …) — ONLINE: fact texts and messages are sent there. See Search by meaning
rerankoffon: rescore the candidate pool before a turn with a cross-encoder (rerank_url, rerank_model, rerank_window, …) — ONLINE: candidate texts and the message are sent there. See Rescoring by a cross-encoder
scopeprofileuser: a separate database per user (holographic-users/<platform>-<user id>.db next to db_path); in a shared chat the store follows the author of each turn, and with no user identity nothing is recalled or written. chat: a separate database per chat (holographic-chats/<platform>-<chat id>.db) — for one bot serving many unrelated groups; thread: per forum topic / thread (…-t<thread id>.db), the chat's own database outside threads. Without a chat (CLI, cron, subagents) nothing is recalled or written. File names: an id made only of A-Z a-z 0-9 _ . - is used as is (Telegram's -1001234567890); any other character becomes _, the name is cut to 60 characters, and -<first 10 hex of sha256(the original "platform-id")> is appended, so distinct ids never share a file

A number that is not one (hrr_dim: abc, min_trust_threshold: high), and a value outside a key's fixed set
(scope: users, recall_source: full, embeddings: maybe, auto_extract: sure; 0.17.8 — before, it was read as the
default, and a mistyped scope put everyone in one store) turn memory off for the session: the system prompt and every
fact_store call name the key and what it must be, never the value, and no database is opened or written until it is
fixed. The embedding_* numbers are replaced by their default, or brought into their range, with a warning in the
gateway log that names the key. A config.yaml the core cannot parse: Hermes serves the copy it last read whole, and
the plugin uses its settings from it with a warning; with no such copy (or a core that raises) nothing tells what
scope was, so memory is off until the file reads (0.17.8; it used to run on the defaults). Warnings give the error's
type only: a YAML error quotes the line it stopped at, which may hold a key.

Optional task outcomes and recorded explanations (separate from trust, off by default):
docs/experience.md.

Two questions in one message (experimental)

«что Злате нужно взять с собой на плавание в четверг и какой чай она любит?» asks two things; the fact about tea holds
every word of its question and still covers too little of the whole message to be recalled. With
prefetch_clause_coverage: on (off by default) a fact the coverage gate left out gets a second look against each of
the two questions alone — only when the message clearly splits into two Russian or English questions, the fact holds
at least two words of its question, and, for a question about a person, that person is one entity the store knows and
the fact names them. A rescued fact never takes the place of one that would be recalled anyway, with or without the
reranker. On a synthetic set it brought both answers into the top five in 15 of 16 two-question messages instead of 12, and changed
no answer on about 620 older test questions; it has not been checked on real conversations. What it misses and every
risk: docs/two-topics.md.

Tools

ToolDescription
fact_storeadd, search, probe, related, reason, contradict, change, end, correct, revert, merge, forget, history, update, remove, list, stats, export, review, entities, alias, rename, delete_entity, split, rebuild_entities, enrich
fact_feedbackRate a fact helpful or unhelpful (moves its trust score), or record that the person confirmed it (0.15.0)
  • search: keyword candidates in any word form (Маши finds Маша, deploys finds deploy, kucingnya finds
    kucing; see Languages), reranked by word overlap and HRR similarity, weighted by trust.
  • Before each turn the provider recalls up to five facts: search candidates that pass a relevance gate (the fact
    must cover enough of the message's informative words; "ok, thanks" recalls nothing) plus the top facts of entities
    the message names. A fact already injected in this session is not repeated for prefetch_repeat_turns turns.
    Facts are framed as data (#id [trust] text), one line each. prefetch_max_tokens caps the block in approximate
    tokens by leaving whole facts out (docs/context-budget.md).
  • add takes entities (who the fact is about; needed for one-word names such as a first name, so later facts
    mentioning it in any form link to it), keywords (see Synonyms) and
    valid_from (the date it started to hold, when the message says). A name in entities that cannot be one (a
    path, a URL, key=value, a single letter, not a string) is not linked, and the answer of add, update, change and
    correct lists it under entities_not_linked by its position in the list (index, from 0) with the reason, not
    the name itself (a refused name may be KEY=secret); the fact is stored all the same. entities given with a new
    text (change, correct, update) replace the names declared on the fact before; without them those stay, but for a
    name only the old text gave; revert brings back the names of the version it restores (see
    Who a new text is about). Its answer lists similar facts,
    possible_conflicts (same subject, a different claim) and past_versions; the agent decides. Text that is not a
    fact (JSON, tool or chat-role markup, instruction-like phrasing, a credential with its value or a token whose shape
    alone says it is one — an API key prefix, a Telegram bot token (also inside a Bot API address), a JWT, a private key
    (0.14.5) —; instructions and credentials in all eighteen languages of Languages, also glued
    to Chinese text (0.15.1); an environment variable that holds a key, a token or a password, by its name — any in
    capitals ending like …_KEY or …_TOKEN and 310 popular ones of model and service providers — and an address
    with a password in it (0.15.2) —, an image (markdown or
    HTML), invisible characters, long blobs, >1000 characters) is refused with the reason. A name that begins with a
    credential word joined to the rest by a hyphen is no credential when the whole word names a thing (0.17.2, issue
    #6): a file with a common extension (token-saver-stats.jsonl, plugins/token-saver/filters/web.yaml), a host of
    three labels and more (password-reset.example.com), a package with a version after == or @
    (token-saver==1.2, token-saver@1.2.3); a file name shaped like a password is none (password-Secret2026.conf,
    password-P@ssw0rd.txt: a capital with a digit, or "@"), nor is an e-mail address. A credential word that stands alone is followed by its value,
    whatever it looks like: password Summer2026.txt, password: secret/hunter2.txt, Пароль файла: Лето2026.txt,
    password-wifi: Qwerty123, Password wifi: correct-horse, and the token counter.py module and Token file: tokens.json too; so is a dot that is
    no extension (password-Qwerty.123), a one-letter or a key file's extension (token-Secret.key), a two-label host
    (password-letmein.co), a tag after : (password-wifi:192.168.1.1, and so token-service:1.4). The cost,
    chosen: password-hunter2.txt is not caught. A request to send a credential stays refused whatever follows the word ("send credentials-backup.json",
    and so "share the token-saver report" too); a Russian credential word counts as a word, not as the start of one
    (секретарю, ключевые цифры, учётная политика, токенизатор, "пин на карте" are none; пин-код, "пин от
    карты", секретная фраза, фраза восстановления are). Recall
    checks a stored fact again before it goes into a prompt (0.13.0): one an older release, or another program, wrote
    past today's rules (markup, instructions, credentials, invisible characters, images) is left out of recall
    and logged once; search, list and history still show it, so it can be changed or forgotten. A quoted
    instruction to the assistant (a note quoted as telling it to drop its rules) may be stored as a report of what
    something says, but recall leaves it out too (0.14.3): in the prompt it reads the same as the bare instruction.
  • stats (what is stored and how much of it was ever recalled; recall_error while recall before a turn fails, so a
    broken store does not pass for one with nothing relevant), export (Markdown view), review
    (clean-up candidates: low trust, never recalled for 60 days, near-duplicates), enrich (facts without keywords,
    most recalled first, to give keywords with update).
  • entities lists names with their aliases, fact counts and standing (0.16.0): established when someone named
    it (entities=[...], alias, rename) or two or more facts name it, provisional when the rules below guessed it
    from one fact and nobody confirmed it; status lists one kind, and the answer says how many provisional names the
    store holds. A name whose letters have no case (Chinese, Japanese, Thai, Arabic...) that the facts only ever put in plain quotes stays
    provisional however many facts quote it (0.16.2): in Chinese a button or a step goes in quotes as readily as a name.
    Then it takes a person naming it, or two facts of which one uses it in title marks 《》, in "X aka Y" or in
    running text where mentions are found (a name of several words, or one a person declared: a one-word Chinese name
    nobody declared is not looked up in running text, where a quoted label is written bare as often). alias gives an entity a second name (or merges two entities; entity stays the name).
    rename gives it another name (the old one becomes an alias; a name another entity has is refused, so a typo
    never merges two). delete_entity removes a name that is not one — a phrase, a command, a label — with its aliases and
    links (entities: [...] for up to 100 in one call, 0.16.2); every fact stays, a tombstone (a keyed hash, not the name) keeps text from creating it again until someone
    names it explicitly, and the file is rewritten as after forget. An alias "X aka Y" in a fact's text gave goes with
    the last fact that says it, and so does X on the facts only that alias linked; an alias written before 0.16.0 is of
    unknown origin, kept like a given one and dropped only when a forgotten fact said it. A name with a comma is no
    name (the legacy aliases column is comma-separated). review lists provisional names older than 60 days under
    entity_candidates.
    Spelling variants are one entity (Hermes Agent = hermes-agent), and X aka Y makes Y an alias.
  • An alias a fact's text gave is a hint, not a decision (0.17.0). "X aka Y" lets later texts that say Y link to X,
    and probe Y find X's facts — but no action someone takes with a name is resolved through it: a name declared
    (entities=[...] in add, update, change, correct) that is only such an alias becomes an entity of its own, and the
    answer says so under split, with the alias call that makes them one again; alias, rename and delete_entity
    by it are refused, and so is rename of someone else into it (its hint would have merged the two). Deleting X
    drops such aliases without keeping Y from being found again (text_aliases_dropped). Until 0.16.3, "Alice Smith aka Bob Brown" after facts about Alice made a later
    entities=["Bob Brown"] file the second person's facts under her, and rename or delete_entity by "Bob Brown"
    renamed or deleted her. alias {entity: X, content: Y} confirms such a name; split {entity: <an alias>} makes
    any alias an entity of its own. Facts whose text names Y then go to Y; one whose text names neither stays with X if
    someone declared X on it, and the answer lists it under facts_to_check — why a declared link exists the file does
    not record. If someone gave X the alias (alias), X stays an entity after the split even without facts; otherwise
    the answer says from_dropped when X went with its last fact. entities shows such names as text_aliases,
    review lists them under alias_candidates however old (up to its limit), stats gives their number. What this
    does not do: a text that says Y still links to X, so recall can bring X's facts for a message about Y until someone
    splits them; it guards what this version writes — an older version writing the same file still resolves a declared
    name through such an alias; and an alias written before 0.16.0 is of unknown origin even if a text gave it: it is
    shown under aliases, not refused, and split separates it. A name declared with a text already stored (an
    add of the same text, a correct of the date only, a change into another fact's text) now counts too; until
    0.16.3 it was dropped.
  • probe: every fact linked to a named entity, from the entity index (name or alias, any letter case).
    reason: facts linked to all of several entities; when a name is not a known entity, keyword search that still
    requires every name. Entities are linked when a fact is stored: capitalized names of two to five words in any cased
    script on one line, ending where the script changes (Alice Smith, Мария Иванова, Jean-Luc Picard; "Docker
    Hub直连可用" gives Docker Hub), short quoted terms ("Phoenix", «Сбер Бизнес», “华为”, 「华为」, 《平凡的世界》,
    straight quotes beside Chinese characters too) and both sides of X aka Y (Y becomes X's alias only while no fact has named Y on its own);
    code spans and blocks give none, and a known name of several words is mentioned only as written together on one
    line (in Chinese, never across a seam between two words); a name begins with a letter or a digit, holds no invisible
    character, passes these checks in its NFKC form too, and flags, key=value, paths, URLs, IPs and commas are not names, and
    neither is a version number with nothing beside it that names a thing ("0.17.0 installed"; "Python 3.12" is one), a
    quoted span whose letters with case are all lower case and that holds no identifier with a letter in it ('count',
    '14 days', «быстро беги»; 'kanban.db' and 'gpt-4o' are), a quoted greeting ("Good morning", «Спасибо большое»,
    “谢谢”), a count ("3个月", "15分钟", "1 小时": an Arabic number and units; 360安全卫士 and 115网盘 are names),
    a quoted Chinese span that reads as something said — ending in a sentence-final particle (“知道了”) or with
    three or more function words besides 的 (“先去搜索再告诉我答案”; titles and brands with one or two, 平凡的世界,
    饿了么, 有赞科技有限公司, are names) — or one of more than 16 Han characters. A name someone gives (entities,
    alias, rename) skips these rules and needs only a letter ("3M", "C++"). A one-word name is not linked; probing it
    falls back to keyword search. When these rules change, a writer rebuilds the index on its first open (up to 2,000
    facts and 500,000 characters; a larger store says so in stats and is rebuilt with rebuild_entities), reading
    the facts in the order they were stored as adding them would; declared names, aliases someone gave and their links
    stay as they are.
  • related: facts structurally tied to an entity in their HRR vector, above a noise floor.
  • Every recall action honours min_trust (default min_trust_threshold).
  • contradict: candidate conflicts, i.e. pairs about the same subject that make a similar but not identical
    claim ("lives in Paris" / "lives in London"). It suggests; it never edits.

History, sources and forgetting (0.8.0)

A fact that stops being current is not overwritten. The agent says what happened, and the store keeps the history:

actionwheneffect
change {fact_id, content, valid_from?, keywords?}it changed (moved, new job, new price)the old text becomes a past version of the same fact id; the old text's keywords go unless given again (the result names those dropped)
end {fact_id, valid_to?}it no longer holds and nothing replaces itthe fact leaves the current store; its history stays
correct {fact_id, content?, valid_from?, keywords?}it was wrongthe wrong version is never shown as the past; keywords given replace the fact's, otherwise they stay
revert {fact_id}undo the last change, correction or endthe previous version is current again
merge {fact_id, from, keywords?, expect?}two or more facts say the same thing (0.19.0)the copies leave the present into the keep; each keeps its history under its own id
forget {fact_id}the user asks to forget itdeleted for good, every version included; only a tombstone without text stays
  • Ordinary recall sees only the present. Past versions live in their own table and index; search and list
    read them only with as_of=YYYY-MM-DD ("what held on that date"). The answer separates results (known to hold
    then) from undetermined (may have held: a start or an end that is not known to the day). The store invents no
    dates: a fact without valid_from is known to hold from the day it was recorded, and so is every fact stored
    before 0.8.0.
  • history {fact_id} lists the versions with their dates, status and source.
  • Source. Each version records where it was written: via (tool, auto_extract, mirror), platform, chat
    type, chat and thread id, session, and the author of the turn. Chat names are not stored.
  • Forget means gone. The fact, its versions, links and orphaned entities are deleted, both search indexes are
    rebuilt, the file is rewritten (VACUUM, with secure_delete on) and the write-ahead log emptied, so no copy of the
    text stays in free pages. A tombstone (the id and keyed hashes, never the text) stops auto-extraction from bringing
    the fact back from old messages and tells other tools which ids to clean from their own copies.
  • Merge, not forget, for duplicates (0.19.0). merge {fact_id: <keep>, from: [ids]} closes each copy's current
    version as merged: the copy leaves search, recall and list; its versions, sources and confirmations stay under
    its own id (history {keep} shows them under merged_from, history {copy} says merged_into), and its last text
    answers exists with the keep when added again, also from auto-extraction — while the keep is still at the
    version it had at the merge (into_version_id); once the keep is changed or ended, the copy's text is its past
    claim (past_version); after a change, end or correction of the keep auto-extraction still skips it. The keep
    keeps its own text, trust and origin; keywords are joined (or given), the counts added once.
    expect: {"<id>": <version_id>} merges only if every named fact is still at that version (version_id is in
    history), otherwise changed_since and nothing is written, so a merge a person approved merges exactly the
    texts they saw. A scheduled job merges only facts it stored in the same session. There is no unmerge; forget of
    the keep forgets its copies too. The first merge raises the history schema (see
    Database and downgrades).
  • Older prompts keep working: add supersedes=X is change X (the id now stays X), update content= is correct,
    remove is forget.

Who stands behind a fact (0.15.0)

The source says where a fact was written; it did not say whether the person said it or the agent concluded it. Six
months on, a rule the user set and a guess the agent made looked the same.

  • Origin. add, change and correct take origin: said (the person said it), confirmed (the agent's
    conclusion, and the person agreed) or inferred (the agent's own conclusion). It is kept with the version, shown
    by history, list, export and search, and counted in stats. Auto-extracted sentences are said (they are
    the user's own words); facts stored before 0.15.0 and mirrored memory entries have no origin. A correct that
    names no origin keeps the version's (fixing a typo does not make a fact nobody's); a change is a new claim and
    takes only the origin it is given. A session opened before an update keeps the tools it was given and offers no
    origin until a new session starts: its facts land without one. A new fact or a changed version stored with no
    origin has none (unknown in stats), and the tool description says so; an add of a text already stored
    keeps that fact's origin and trust. In the database the origin is kept with
    the version, in fact_versions.src_origin of the current row (status = 'current'), not on facts. Stored
    times are UTC (fact_versions.recorded_at, facts.created_at / updated_at); for a window in local days use
    fact_versions.recorded_on, the same moment as a date in Hermes's timezone (UTC when none is set).
  • Starting trust. What the person said or confirmed starts 0.1 above default_trust, the agent's conclusion 0.1
    below it (within 0 and 1), a fact without an origin at default_trust as before; a change starts over at its
    new origin's trust, and so does a new text through correct or update (its origin is the version's unless the
    call names one), without the old text's helpful marks and recall count; a new date alone keeps them. Measured on 40 pairs of conflicting facts about the
    same person (one said, one inferred, stored in alternating order): without origins the one said came first in 17;
    with a step of 0.05, 0.1 or 0.15, in 40 of 40, in search and before a turn alike. An inference with no rival
    was still recalled every time (10 of 10). 0.1 keeps the order after a helpful vote or two on the inference.
  • Confirmations. fact_feedback {action: confirmed, fact_id} records that the person confirmed a fact — said it
    again or agreed when asked — and the same text added again with origin: said in a later session counts too (not
    in the session that stored it: that is the agent repeating itself). At most one confirmation per fact per session,
    or per day where there is no session. Trust rises by 0.05 for each person who confirms a fact the first time:
    one group-chat member saying "yes, true" session after session raises it once (without a known id, once per
    session). An inference that has been confirmed counts as confirmed; a change drops the confirmations, which
    were of the old claim, and so does a new text through correct or update. A nightly job that finds a person's message backing a fact can confirm it the same way.
  • Recall says what is a guess. Before a turn, an inference nobody confirmed ends with (inferred), so the model
    does not repeat it as something it was told. review lists inferences nobody confirmed in 60 days.

On a family store the agent stored 22 facts in four months and never the same one twice, so confirmation by
repetition alone will be rare there: the explicit confirmed and a nightly job are where confirmations come from.

The full contract, including what the plugin cannot clean (Hermes's own session database, hermes backup
archives): docs/history.md.

Nightly dreaming: hermes-dreaming

The store keeps what the agent was told; it does not decide what deserves to become a permanent memory.
hermes-dreaming (in the Hermes plugin catalog as
hermes-dreaming) does that at night, deterministically and without its own model calls: an optional extraction job
stores candidate facts here through fact_store, and the nightly pass reads this store read-only, corroborates each
fact with what people actually said in their chats, proposes the few worth writing to MEMORY.md / USER.md, asks
whether old entries still hold, and gives facts without keywords their keywords. With 0.8.0 the two work as one: a
human's "no longer true", "that was wrong" or "forget it" to a dream question becomes end, correct or forget
here, and after a forget the dream removes its own copies of the fact (cache, diary lines, its cron logs) by the
tombstone's id. Use hermes-dreaming 2.4.0 or later with this release: cron jobs may write facts only from 0.8.0 on
(cron_writes).

Memory map (experimental)

memory_map.py exports one store as a read-only JSON snapshot a person can read: the facts with who said them and
how much they are trusted, people and topics, contradictions and which side search leans on, the week's changes,
forgotten facts as a number and a date, and what the nightly review asks about this store. The
hermes-dashboard "Memory map" tab draws it. Nothing calls it
unless you do (python <plugin dir>/memory_map.py --store PATH --json); the provider is unchanged by it.

Experimental: the format and the rules may change. Its output is personal data — keep any page made from it
behind a password. What it shows and hides, and its limits (a closed word list for "person", "search leans on" only
for a question that fits both facts equally, the night's files outside the database snapshot, a picture that ages
until the next rebuild, at most 1000 facts and 300 names by default — --max-facts / --max-entities, 0 for no
limit): docs/memory-map.md.

How to see it:

  • In hermes-dashboard (a page behind basic-auth, rebuilt by its cron): turn on its views.memory_map and point
    memory_map.plugin_dir / memory_map.store at this plugin and the store — step by step in the dashboard's README,
    "Memory map (experimental)".
  • Your own page or tool: run the exporter and draw the JSON; every key is described in
    docs/memory-map.md. Draw only what it gives — the privacy rules are
    applied by the exporter, so a renderer must not open the database itself — and escape every text (they are what
    people told the agent).
  • A ready-made standalone HTML file and a fact_store action that builds it are not in this release.

Languages

How a word is matched across languages: words, language, function words, keys, match; examples in eighteen languages
that match (German compounds, Russian and English spellings of one word, Traditional and Simplified Chinese, Japanese
spelling variants) and examples that do not

A word matches another when they share its plain form or its stem in the language of the text:

  • Stems: Snowball for English, Russian, Spanish, French, Portuguese, Indonesian, German, Hindi, Italian, Turkish,
    Polish and Dutch, with two plural rules it lacks (Portuguese ação / ações, Spanish lápiz / lápices); for
    Arabic, Larkey's light stemmer (the one Lucene ships) next to Snowball's, plus the word without a glued conjunction or
    preposition before an alef (لإيرما, "for Irma"; ولإيرما, "and for Irma"); for Bengali, a light suffix stripper
    (Snowball has none) that leaves at least three letters, so মাটি (soil) is not মা (mother). Italian, Turkish,
    Polish and Dutch words (0.11.0) are stemmed as written and folded: Snowball's Turkish and Polish read the letters
    folding drops (çocuğun is çocuk, domów is dom), and a word typed without its marks still meets one typed with
    them. Turkish also gets the noun without the s of a possessive that Snowball leaves (kedisi → kedi); Dutch keeps
    a ge- that Snowball takes off (gezin, family, is not zin, sense). Polish needs snowballstemmer 3.1 or newer
    (current Hermes installs 3.1.1); with 2.x Polish words compare as written and the rest of Polish (function words,
    greetings) works.
  • Which language: the script decides for Cyrillic, Arabic, Devanagari and Bengali. Latin text is English, Spanish,
    French, Portuguese, Indonesian, German, Italian, Turkish, Polish or Dutch by the function words it uses ("the", "el",
    "le", "o", "yang", "der", "il", "bir", "się", "het"). Text with none of them is stemmed in English, Spanish, French,
    Portuguese, Indonesian and German, and in Turkish, Polish or Dutch only when it has their letters (ı ğ ş; ą ę ł ń ś ź ż ć; ij): Polish stems of English words join unrelated ones (army / arm, apply / apple), so a
    one-word Polish or Italian question gets its stems only with a function word or such a letter. A word is dropped
    as a function word only when the other words show that language, or when it is the only sign of it in a sentence of
    four words or more: "son" alone, or in "where does my son study", is a search word; in "il boit son café" it is
    French "his". With no other evidence the English list applies, as it did to all Latin text before 0.5.0. Function
    words of the four newer languages that are also everyday words, names or abbreviations elsewhere (Italian come,
    dove, per; Turkish kim, ben; Polish bo; Dutch door, met, van, wat) do not tell the language and are
    dropped only where that language alone fits the other words: "Ana come carne" stays Spanish, "Kim Lane lives in Oslo"
    keeps kim. Italian mia (my) is not on the list: it is a common name.
  • Folding: Latin accents and ligatures (esta finds está, coeur finds cœur), Arabic vowel marks and letter
    variants, the Indic nukta, and native digits (2027 finds ২০২৭, HR24 finds HR২৪). Folding also merges what
    only an accent tells apart (Spanish año / ano), as the FTS5 index already did. Turkish ı and Polish ł compare
    as i and l (İstanbul, ISTANBUL and istanbul meet, KIZIM meets kızım, szkola meets szkoła); since the
    FTS5 index keeps those two letters, such a word is also searched as written.
  • One word, two scripts (0.14.5): Russian text writes a borrowed word either way, and бэкап finds backup,
    телеграме finds Telegram, впн finds VPN, визы finds visa, and the other way round — as the same word,
    without search by meaning. 55 such words are listed (apps and services, devices, server, backup, token,
    hotel, visa, chat, site, bot…; _LOANWORDS in text.py). A word belongs to its group by its exact form
    — the spelling, an English plural, a Russian case ending — never by its stem: Snowball cuts токен to ток
    (electric current) and виза to виз, as in визит, and neither meets token or visa. A Latin spelling joins
    its group only in English or in text of no known language (0.14.6): French chat is a cat, German über folds to
    uber, Spanish botes are boats. A spelling whose Russian forms are another word is not listed (убер gives
    уберу, of убрать), and a short one is searched by its forms as whole words, so бот does not bring in
    ботинки. Words two languages share by meaning, not spelling (password / пароль), are not listed: that is what
    keywords are for.
  • Hyphens: French elision and inversion (d'Irma, prend-il, est-ce) and Portuguese enclisis (chama-se) are
    taken apart in text of that language, reduplication (anak-anak) in any; a hyphen between scripts splits
    (hr-vps-এর). A letters-only hyphenated token is an identifier in English text or text of no determined language
    (hr-vps, carry-on): a message naming it is about that one, and a fact about hr-other is not recalled for it.
    In Spanish, French, Portuguese, Indonesian and German text it is a compound word (week-end, sexta-feira,
    IT-Treffen), so a host name in a question in those languages is matched like any word, not enforced.
  • Apostrophes: Italian elision as French (dell'anno, l'ospedale, quest'anno, dov'è, anch'io; not before
    a lone s, so English dev's stays). A Turkish case ending after a name comes off in text that may be Turkish
    (Ankara'da, Ayşe'nin, HR24'ün, hr-vps'de are Ankara, Ayşe, HR24, hr-vps), so the name is mentioned and an
    identifier stays required; only Turkish endings, so Irma's, o'clock and rock'n'roll are left alone. In Dutch
    text the plural and possessive 's comes off (pinda's).
  • German compounds (0.11.0): in text determined as German, a compound is split by a dictionary of German lexemes
    from Wikidata (CC0): parts of three letters or more joined by German linking elements (-s-, -es-, -n-, -en-,
    -er-, -e-, a dropped -e), the analysis with the fewest parts, Wikidata's own analysis where it has one
    (Zahnarzttermin = Zahn + Arzt + Termin, Blumentopferde = Blumen + Topf + Erde); two analyses that cut
    differently leave the word whole unless one is clearly likelier (Staubecken, Wachstube stay whole). Both
    directions: the last part of a compound in a fact matches like the word itself (Katze finds Familienkatze), and
    a compound in a question finds a fact that writes its parts apart (Zahnarzttermin finds Termin beim Zahnarzt).
    The parts before the last count 0.6 and never recall a fact alone, and so does a last part of three letters
    (Tag in Alltag); a question compound whose last part alone is in a fact counts 0.6 too. Words the dictionary knows as one word stay whole (Matrose is no Rose, Mitglied no
    Lied, Ausland no Land); 0.6.0's suffix test stays for words it does not know (Arbeitslaptop). On Wikidata's
    own compounds held out of the build, 94.9 % are split, 99.9 % of the splits as Wikidata cuts them; of 204 simple
    words that look splittable, 4 are still split (Plattform, Tablette, Kommode, Zoophobie). Compounds whose
    meaning is not their parts' are never split (0.12.0: Hochzeit is no Zeit, Frühstück no Stück,
    Weihnachten no Nacht, Augenblick, Verstand, Reisende, Wochenende, Flugzeug …, nor their forms;
    Tablette is in that list too), in the table as well since its rebuild (0.17.0): the build keeps those words
    whole and cuts the other nouns as before, by Wikidata's analysis where it has one and else by the splitter's own
    judgement with stricter limits (0.12.0 had made every noun without an analysis one word; the table it would have
    built no longer split 15 of 33 everyday compounds, and was never shipped). The table of 0.17.0 was built without
    the verb forms and lacked 1,530 verbs that keep their forms whole: it cut 1,515 forms (gehörte as geh|orte,
    stelltest as stell|test); 0.17.1 is built with them, and verbs, adjectives and verb forms are read as in 0.16.3
    again. A real ambiguity
    stays: Kopfschuppen (dandruff) is Kopf + Schuppen, and Schuppen is also a shed. de_compounds.tsv.gz
    (212 KB, rebuilt 2026-10-17), loaded on the first German word (0.15 s, 7 MB); _DE_COMPOUNDS = False turns it off.
  • German and Spanish verb forms (0.11.0): Snowball keeps ging apart from gehen, fue from ir and ser,
    hizo from hacer. A table of such forms from Wikidata's lexemes (the 2,000 German and 4,000 Spanish verbs a
    stand-in for frequency ranks first) gives a form its lemmas' stems, in text of that language only (not in a tie
    with English: "he went"). Such a match counts 0.6, like a root, and never recalls a fact alone: a form can belong to
    two verbs (fue is ir and ser), the source has errors (tuviese under twelve verbs, dropped: a form under
    more than three is left out; gelegt under liegen), and a name can be spelled like a form (Lucía / lucir).
    Forms spelled like another word are left out (Macht, vino, cuenta, hecho), and so are forms spelled like a
    regular form of another verb (fuhren / führen). Separable verbs are not joined (rief … an / anrufen). German
    text is told by its function words, and the list lacks zum, ins, wohin: a short German sentence without others
    stays undetermined and gets neither compounds nor verb forms. de_es_forms.tsv.gz (251 KB): German 0.03 s and 2 MB, Spanish 0.05–0.09 s and
    11 MB on the first word of the language; _LEMMA_FORMS = False turns it off.
  • Dutch compounds are not taken apart (_NL_COMPOUNDS, off): on 10 questions that ask by the end of
    a compound and 7 that use that end in its own sense, it found 2 more right facts and recalled 4 more wrong ones
    (huis in ziekenhuis, weer in brandweer, kast in koelkast).
  • Words of the same root (Russian and English): Snowball removes inflection but keeps derivations apart. Russian
    pairs come from a dictionary (0.10.0): DerivBase.Ru, its direct derivation links as stem pairs, kept when the pair
    has the shape of a derivation (ночные / ночь, городской / город, учительница / учитель, дружеский /
    друг; not полный / пол or задача / дача) — 46,756 pairs, Apache-2.0
    (licence), loaded on the first Russian word (about 0.3 s, 12 MB). English
    pairs come from Open English Wordnet 2025 (0.11.0): its derivation and pertainym links between single words, as stem
    pairs of the same shape (teacher / teaches, pay / payment, deliver / delivery, decide / decision,
    angry / anger, allergic / allergy; not Peter / pet, meters / met or classic / class, which
    0.9.0's suffix rules joined). A link is kept only when it is the derived word's commonest sense (butter is food
    before it is one who butts, tower a building before one who tows, letter writing before one who lets) and no
    stem is another, unrelated word (pearly stems to pear) — 5,428 pairs, CC BY 4.0 on Princeton WordNet 3.1
    (licence, WordNet notice), 30 KB, loaded on the
    first English word (0.04 s, 2.4 MB). Missing from it: dentist / dental, anxiety / anxious, runner /
    run, rental / rent, medical / medicine (left out by those filters); pronunciation / pronounce is in
    it but its spelling change is too large for the shape filter. _EN_DERIVATION = "rules" (and _RU_DERIVATION) restores 0.9.0's rules. Russian
    stems of six letters also match on their first five (покупа / покупк). Only in text whose language is
    determined (English by its function words; Cyrillic without Ukrainian or Belarusian letters), and only within that
    language. A root counts 0.6 of the word itself and never recalls a fact on its own: suffix rules cannot tell a
    derivation from a look-alike (войной / вой, полный / пол), so automatic recall needs a word of the message
    itself, and roots add to it. In search a root brings in facts the word's forms missed and ranks them below facts
    with the word. Not joined: alternating consonants and prefixes (друг / дружба, страховка / застрахован),
    four-letter beginnings (большое is not больница). Duplicate and conflict checks do not use roots.
  • Courtesy (0.9.0): a message whose every clause is greetings, thanks, sign-offs, reactions or "how are you",
    at least one of them more than padding, recalls nothing ("Hi, good morning!", "Thanks, that's all for now",
    "спасибо, на этом всё", "Grazie mille, a domani!", "Wielkie dzięki, do jutra!", "Heel erg bedankt, tot morgen!",
    and any combination of the whole-message greetings of every language). Pieces start and end
    at word boundaries; padding alone ("fine?", "сегодня?") is a question. Russian question words (какой, который,
    сколько, почему …) are left out of a query's terms, like English what / which / how, but kept in facts.
  • Names in Latin script are matched as written (and with an English 's): Jean is not mentioned by jeans.
    Names in other scripts carry their inflections (Маши mentions Маша, لإيرما mentions إيرما).
  • Chinese, Japanese, Korean, Thai (no spaces between words): character n-grams, bigrams for Chinese, Japanese and
    Korean, trigrams for Thai. Grammar weighs less: Chinese function characters (的 是 了), Japanese particles (は が を
    に の), Thai function words (ของ ที่ และ); a message that is only a filler (はい, 네, โอเค) recalls nothing. Chinese
    characters (also in Japanese) count one by one as well; kana, Hangul and Thai letters, which are sounds, do not.
    Traditional characters compare as Simplified ones (0.9.0): 學校 finds 学校, 醫院 finds 医院, so a fact a Taiwan
    or Hong Kong user wrote meets a mainland user's question and the other way round; the stored text keeps its script.
    The table is OpenCC's (cjk_fold.py, 3,222 characters, Apache-2.0, licence). The
    n-gram scan reads the facts' folded text, analysed once per fact: at 5,000 Chinese facts the first search after start
    takes about 1.5 s, later ones 0.12 s with recall (0.42 s in 0.8.0, whose scan ran in SQL).
  • Japanese spelling variants (0.11.0): one word written in kanji or kana, with more or less okurigana, or as
    a loanword with or without the long vowel compares as its representative spelling: 子ども finds 子供, 引越し and
    引越 find 引っ越し, 申込 finds 申し込み, コンピュータ finds コンピューター, ウィルス finds ウイルス. Only in text
    that has kana, so Chinese text is never touched; the stored text keeps its spelling. The spellings come from
    SudachiDict's synonym source (ja_variants.tsv.gz, 5,167 variants of 4,011 words, Apache-2.0,
    licence), spellings of one word only: no synonyms, abbreviations (スマホ is
    not スマートフォン), translations or romaji. Japanese has no spaces, so a variant is replaced only where its edges can
    be word edges: hiragana next to a particle, punctuation, katakana or a suffix (たち, さん), not inside other kana
    or right after a kanji (べんきょう holds no きょう, 白い草 no い草); a short katakana word not inside a longer one
    (アイス holds no イス). Left out: hiragana of two letters (いす, かに), hiragana verbs Sudachi marks as ambiguous
    (はなす: 話す, 離す, 放す), words whose representative is one kanji (たまご / 卵: a lone kanji has no bigram, so
    two facts that both wrote たまご would stop meeting), and いつも / だいぶ (何時, 大分 are other words). Verbs
    meet in the forms the source lists and through their noun (引越しします). Loaded on the first text with kana:
    0.03 s, 2 MB.
  • Chinese words (0.11.0): character pairs cannot tell a word from the halves of two words, so text whose Han is
    Chinese (no kana, no Hangul) is also split into words, the way jieba cuts without its HMM: the most probable split
    by word frequency, a character that begins no word standing alone (南京市长江大桥 is 南京市 | 长江大桥, 我和服务员
    is 我 | 和 | 服务员). The words add to the pairs, they do not replace them. Where two dictionary words meet, or one
    meets a function character standing alone, the pair across that seam is read as such: a question's 和服 (kimono)
    is not found in 我和服务员 ("me and the waiter"), 研究生 (graduate student) not in 研究生命 ("to research life"),
    校长 (principal) not in 学校长期, and a character a fact holds only as the half of such a split pair does not count
    one by one either. Next to a content character standing alone the split is not sure enough: 看 | 牙医 still holds
    看牙, 续 | 保险 still 续保. Every pair keeps its weight, so the share of a question a fact covers is counted as
    before. A one-character word is also searched alone, wherever it stands in a fact: 猫几岁了? finds 橘猫,
    药一天吃几次? finds 降压药; not function characters, numerals, units of the calendar, the commonest words (过,
    看, 去), characters jieba tags as a person name (most surnames), or two side by side in a question, which are more
    likely a name the dictionary lacks (怡君, 思琪: found by their pair, as before). Traditional text is split after
    folding; Japanese and Korean text is compared as before. The dictionary is jieba's (zh_words.txt.gz, all 347,549
    words of its dict.txt, folded to Simplified, frequencies kept to a quarter of a natural-log step; 1.45 MB; MIT,
    licence; its author has said the frequencies were counted on the People's Daily 1998 and
    MSR segmentation corpora, which carry research terms of their own — the dictionary is used as jieba publishes it);
    the segmenter is the plugin's own, jieba is not a dependency. Loaded on the first
    Chinese text: 0.2 s, 30 MB. Splitting every fact once adds 0.3–0.7 s to the first search after start at 5,000
    facts; later searches are unchanged within measurement noise. _ZH_WORDS = False in text.py turns it off.
  • Arabic, Indic scripts, Indonesian, German, Turkish, Polish and Dutch are also matched word by word against the
    text of the facts (so are Latin words of a question of undetermined language, when the store holds facts in one of
    those Latin languages): the SQLite FTS5 index splits Indic words at their vowel signs and cannot see past Arabic and
    Indonesian prefixes, into German compounds, through Turkish consonant changes (kitap / kitabı) and Dutch plurals
    that change the word (raam / ramen), or from szkola typed without its ł to szkoła. The rows are read on every such search; a fact is analysed once and again only when its text changes.
    Measured at 5,000 facts: 0.03–0.07 s per search and about the same for the relevance gate; the first such search
    after start analyses every fact, about 1 s at 5,000 facts and 4–6 s at 20,000 facts (later searches 0.2–0.4 s, plus
    0.15–0.2 s for the gate). A store without such facts never scans: an English store answers in about 0.02 s at 5,000
    facts, as before. A Turkish store of 5,000 facts (0.11.0): the first search after start 1.5 s, later search and
    recall together 0.25 s (0.11.0 scans it; 0.10.0 answered in 0.01 s and missed most other word forms). Whether a
    store holds such facts is told by a few common words of those languages; a question itself told to be in one of
    them scans in any case, so a store of short facts without those words is still read word by word. Not covered: a
    one-word question of no told language (szkola, typed without its ł) against such a store, and a short fact with
    no function word and none of its language's letters (Ramen gesloten.), which gets no Dutch, Polish, Turkish or
    Italian stems.

Measured on stores of 20 everyday facts per language (people, school, work, servers, plans). "Found" = the right
fact among the first five search results; "recalled" = injected before the turn; "quiet" = chit-chat, greetings and
off-topic questions recall nothing. No keywords in these stores.

Main set (20 questions): found / recalledUnseen set (8 questions, 15 unrelated facts added): found / recalledQuiet (all sets)
English100 % / 100 %100 % / 100 %14 / 14
Russian100 % / 100 %100 % / 88 % ¹14 / 14
Chinese100 % / 100 %100 % / 100 %14 / 14
Spanish100 % / 100 %100 % / 100 %10 / 10
Hindi100 % / 100 %100 % / 100 %9 / 10
Arabic100 % / 100 %100 % / 100 %10 / 10
French100 % / 100 %100 % / 100 %10 / 10
Bengali100 % / 100 %100 % / 100 %10 / 10
Portuguese100 % / 100 %100 % / 100 %10 / 10
Indonesian100 % / 100 %100 % / 100 %10 / 10
Japanese95 % / 95 %100 % / 100 %10 / 10
Korean100 % / 100 %100 % / 100 %10 / 10
Thai100 % / 95 %100 % / 100 %10 / 10
German100 % / 100 %75 % / 75 %10 / 10
Italian ²100 % / 100 %—5 / 5
Turkish ²100 % / 100 %—5 / 5
Polish ²100 % / 100 %—4 / 5
Dutch ²100 % / 100 %—5 / 5

The right fact is first in 95–100 % of main-set questions and 62–100 % of unseen ones. Recall before the turn is not
only the right fact: about one other fact of the store comes with it per question (0.8–1.2 on the main sets), and in
the unseen sets 0–2 of the 15 unrelated facts per language. English, Russian and Chinese also have a second set (10
questions): found 90 / 90 / 100 %, recalled 80 / 90 / 90 %. ² Italian, Turkish, Polish and Dutch (0.11.0): the main
set was written with the code; their blind sets are below.

How the numbers were obtained, honestly:

  • Main sets are tuned on. English, Russian and Chinese: the fixes in 0.4.2 were made against them. Spanish, Hindi,
    Arabic, French, Bengali, Portuguese, Indonesian: first run on 0.4.2 (found 85–100 %, recalled 50–100 %, Arabic 50 %,
    French 75 %, Spanish 85 %, chit-chat recalled facts in four languages), fixed in 0.5.0. Japanese, Korean, Thai,
    German: first run on 0.5.0 (found 95–100 %, recalled 95–100 %, but chit-chat recalled facts in Korean 1/5, German
    1/5 and Thai 4/5), fixed in 0.6.0. The fixes are per language and per script, not per question.
  • Unseen sets were written before those fixes were seen to work and run once after them. English, Russian,
    Chinese: after the run, whole-message greetings were made to recall nothing ("good morning" recalled facts about
    mornings), and the quiet column includes that fix. Spanish to Indonesian: first run before the review fixes below,
    with the same found and recalled numbers. Japanese, Korean, Thai, German: run once; nothing changed after.
  • An independent review (Codex CLI) of 0.4.2 and 0.5.0 found regressions the sets did not: a lone "son" dropped as
    a Spanish function word, hr-vps no longer enforced, ação / ações apart, "with" / "without" sugar no longer a
    conflict, Jean mentioned by jeans, and others. Each is fixed with a test, and every set was run again. One of
    those fixes made French est-ce a required identifier, which the unseen set showed (French recalled 88 %); the
    remedy (compound words in Romance-language text, a hyphenated word votes for a language with its first part) was
    made after seeing it, so that unseen-set number is no longer blind.
  • The second set was written after the main set of 0.4.2, first run without changes (Russian recall 90 %, Chinese
    70 %), and two fixes followed. Its numbers published with 0.4.2 (recalled 90 / 100 / 100 %) were partly luck: facts
    of a named person saved in the same second came back in arbitrary order. They now come back newest first, every
    time, and the numbers above are what that gives.
  • Words of the same root and courtesy (0.9.0) were measured on blind sets written by separate agents that never
    saw the code. Same-root words, two sets each for Russian and English (24 facts, 20 unrelated facts that share
    beginnings or roots with them, 20 questions — 12 asked with another word of the fact's root, 4 with another form, 4
    with a synonym — and 10 traps). The first set was run once before and once after; its misses were read and three
    faults fixed, so it no longer counts. On the second: Russian first 45 → 55 %, found 55 → 60 %, recalled 55 → 55 %;
    English first 60 → 60 %, found 70 → 75 %, recalled 60 → 60 %. Then an independent review (Codex CLI) found
    look-alike joins that these sets had no trap for (войной / вой, mothers / moth, Spanish comer / come);
    since then a root never recalls on its own, roots stay within a determined language, and the second set gives the
    same found and recalled numbers. A second review found padding-only courtesy ("fine?") and question words lost from
    duplicate checks (fixed) and a capital-letter name rule that lost ordinary words ("Baby" / "babies"; removed — on
    every set it changed nothing). Small talk: a fresh set per language (24 facts that use small-talk words, 30
    small-talk messages, 15 questions), run once on the final code: quiet English 50 → 67 %, Russian 63 → 70 %,
    questions found 100 % before and after; what still recalls is a word in another sense or a famous name ("Peter
    Pan", "Лев Толстой", "скрипичный ключ"), which matching words cannot tell apart. Every earlier set gives the 0.8.0
    numbers, except ¹ the Russian unseen set: its one miss ("во сколько ночные бэкапы", nightly / night) is now found,
    not recalled (a root alone); that question is the one the change was written for, so this number is not blind.
    Speed at 5,000 facts is unchanged (0.03 s for a search and a recall).
  • Traditional and Simplified Chinese (0.9.0): two blind sets by separate agents (24 facts, half in each script; 20
    unrelated facts; 20 questions — 6 in the fact's script, 6 in the other one, 4 paraphrases, 4 synonyms — and 10 small
    talk). On the first, a question in the other script was found but recalled in 3 of 6; its misses were read and the
    folding written. On the second, run once on the final code: other-script questions found and recalled 33 → 100 %,
    all questions first 55 → 75 %, found 55 → 75 %, recalled 50 → 70 %, small talk quiet 80 % before and after;
    same-script questions 100 % both; synonyms with no shared character 0 % both (keywords cover them). Folding also
    makes a same word in another sense meet across scripts ("圍棋是誰發明的?" recalls the family's Go lessons, written
    围棋), as it always did within one script.
  • Russian dictionary (0.10.0): a fresh blind set (12 same-root questions, 4 other forms, 4 synonyms, 12
    look-alike traps), run once: same-root questions found 75 % (0.8.0) → 83 % (0.9.0 rules) → 92 % (dictionary),
    recalled 42 → 67 → 67 %; traps quiet 11 of 12 in all three. The choice of the dictionary was made on the earlier,
    already read sets, before this run.
  • 0.11.0 was measured on fresh blind sets, one per language, written by separate agents that never saw the code
    and run once on the code of that moment against 0.10.0; every choice (dictionary or rules, which pairs to keep, what
    to leave out) was made before that run, on the sets above or on development sets written with the code. Then an
    independent review (Codex CLI) found false matches the sets had no trap for (below); after its fixes the same sets
    were run again, and those second numbers are no longer blind.
    • English same-root words (Open English Wordnet; 24 facts, 20 look-alike distractors, 12 same-root questions, 4
      other forms, 4 synonyms, 12 look-alike traps): same-root questions found 100 % both, recalled 75 → 83 %; all
      questions recalled 65 → 70 %; traps quiet 12 of 12 both. After the review's filters (commonest sense, no
      unrelated stem) the same-root questions are recalled 75 %, as with 0.10.0's rules: the filter that drops
      butter / butt and letter / let also drops medical / medicine. On the earlier English sets only one
      question moved:
      "When is Oliver's dental checkup?" is still recalled but no longer first (the dictionary has no dentist /
      dental).
    • Japanese spelling variants (24 facts, 20 distractors that share characters or kana with them, 14 questions with
      the word in another spelling, 6 in the same, 4 synonyms, 8 traps): other-spelling questions found 57 → 71 %,
      recalled 29 → 43 %; all questions first 46 → 54 %, found 71 → 79 %, recalled 46 → 54 %; traps quiet 8 of 8 both.
      After the review's fix (a hiragana variant starts after hiragana only when that is a particle after a kanji or
      katakana word: かいちょう, chairperson, held いちょう, ginkgo): other-spelling questions found 64 %, recalled
      36 %.
      Still missed: some hiragana spellings (はいしゃ, りんご), one-kanji words (ネコ / 猫, left out on purpose),
      and questions found but held back by the recall gate.
    • Chinese words (24 facts, 20 distractors, most holding a fact's word across a seam or inside another word; 10
      questions by a shared word, 4 paraphrases, 4 by a one-character word, 4 synonyms, 10 traps): first 59 → 73 %,
      found 64 → 77 %, recalled 41 → 41 %; one-character words found 0 → 50 %, recalled 0 % both; traps quiet 6 of 10
      both. Splitting moves the right fact up and finds one-character words; it does not lift recall, and traps that
      share a real word with a fact (研究 in both 研究生 and 研究生命) still recall it. The two Chinese sets of
      0.9.0 (read since): first 65 → 80 % and 75 → 80 %, recalled 70 → 70 % and 70 → 75 %.
    • Italian, Turkish, Polish, Dutch (each: 22 facts, 18 distractors, 10 questions in another word form, 6 in the
      same, 4 synonyms, 5 small talk, 5 traps), recalled / small talk and traps quiet: Italian 80 → 85 % / 50 → 90 %;
      Turkish 35 → 85 % / 70 → 50 %; Polish 55 → 75 % / 60 → 70 %; Dutch 70 → 85 % / 70 → 60 %. Other-form questions
      recalled: Italian 90 → 100 %, Turkish 10 → 100 %, Polish 50 → 90 %, Dutch 60 → 90 %. Turkish recalled more where
      it should and more where it should not: "Günaydın canlarım!" (good morning, my dears) met a boy named Can
      (canlarım and Can share a stem), and "Tamam, akşam görüşürüz" (see you tonight) was not read as a sign-off;
      terms of endearment and times of day now pad a Turkish courtesy, and the set gives quiet 70 % (not blind).
      Known weak spots of the stemmers: Turkish verbs are not reduced (okuyor / okudu), and yazdı (wrote) meets
      yaz (summer); Polish consonant changes (kotka / kotce) and the moving e (cukier / cukru); Italian
      chiavi → chi; Dutch irregular plurals (kind / kinderen).
    • German compounds and verb forms (22 facts, 18 distractors; 6 compound questions, 10 other forms, 6 same, 4
      synonyms, 10 small talk and traps): first 62 → 69 %, found 73 → 81 %, recalled 65 → 77 %; compound questions
      found and recalled 67 → 83 %, other forms recalled 70 → 90 %; quiet 8 → 7 of 10 (Kopfschuppen, above).
      Spanish (6 irregular-verb questions among 26): no change at all, 73 % recalled before and after — its irregular
      questions shared another word with their facts and were found by 0.10.0 already, so this set says nothing about
      the forms table either way; the table rests on development sets (first 72 → 84 %).
    • Every earlier set gives the 0.10.0 numbers in every language, except the Chinese ones (above), the English
      dental question, and German, whose synonym set's questions are found 25 → 50 % (88 % with keywords); the Russian
      sets are unchanged.
  • Review of 0.11.0 (Codex CLI), each fixed with a test (tests/test_holographic_v0110_review.py): Polish stems in
    text of no determined language joined English words (army / arm; now only with Polish letters); a host name in a
    Turkish, Italian or Dutch question was no longer required (hr-vps'de); a hiragana variant was replaced inside
    another word (かいちょう); English links of a rare sense were applied to the common one (butter / butt, tower
    / tow, pear / pearl); a Chinese one-character word recalled an idiom (猫 / 猫腻; now it needs a pair of the
    question or a second character); a kanji-only Japanese question missed a fact whose kana folded its spelling (a
    fact keeps its spelling as written too); Turkish, Polish and Dutch stores without the probe words were not scanned;
    "Günaydın canlarım!" recalled facts about Can; a damaged data file broke recall instead of being skipped. A second
    review (of German and Spanish, and of these fixes) found verb forms recalling facts alone (tuviese / contiene,
    gelegt / liegt, Lucía / luce, fue / va: now 0.6, never alone), a three-letter compound head doing the
    same (Alltag / Tag), a host name lost in a Dutch tie with Spanish (De server hr-vps: compounds only when every
    tied language writes them), and a malformed compound row raising instead of being skipped; each has a test. The
    German set gives the same numbers after these fixes, quiet 8 of 10 again; every other set is unchanged.
  • Not measured: other languages (exact words, no stems, no function-word list): among them Vietnamese, Ukrainian
    (compared with Russian stems), Korean beyond its n-grams.

Synonyms and other roots: keywords

Words are matched, not meanings: "relocating" does not find "move", 宠物 (pet) does not find 猫 (cat); words of one root
are joined only in Russian and English (see Languages). The plugin calls
no model, but the agent that stores a fact is one: fact_store add takes keywords, 3–8 words someone might ask
with that the fact does not contain (synonyms, other roots, broader terms, the same in the user's other languages).
They are kept with the fact's tags, searched like its text, never shown in the recalled block, and old versions index
them too. This is document expansion, as doc2query does for lexical search: the meaning is added once, when the fact
is written, and recall stays local and deterministic. enrich lists the facts that have none, so the agent or a
nightly job can add them with update; an update with keywords alone keeps the fact's age (updated_at), so old
facts do not turn into recent ones. hermes-dreaming does this
nightly (keywords.enabled).

Measured on a third set per language: 8 questions that share no word with their fact except a name ("Where are we
relocating?", "What breed is our pet?", "Can my office computer handle AI models?"), written after the keywords. The
keywords came from a separate agent that saw only the facts and the description of keywords, not the questions; a
first batch was thrown away because the description's example then was the store's own "we move to Hua Hin".

Without keywords: found / recalledWith keywords: found / recalled
English38 % / 38 %100 % / 100 %
Russian25 % / 25 %75 % / 62 %
Chinese38 % / 12 %88 % / 75 %
Spanish38 % / 25 %100 % / 100 %
French38 % / 38 %100 % / 100 %
Portuguese38 % / 38 %75 % / 75 %
Indonesian38 % / 25 %88 % / 88 %
Arabic38 % / 38 %75 % / 75 %
Hindi38 % / 38 %88 % / 75 %
Bengali38 % / 38 %75 % / 75 %
Japanese50 % / 25 %88 % / 75 %
Korean50 % / 12 %88 % / 62 %
Thai38 % / 25 %75 % / 62 %
German25 % / 38 %62 % / 75 %
All 11242 / 3394 / 88

Without keywords, what is found comes through the name in three of the eight questions (Irma, Anton). With them, the
main sets lose nothing (280 / 280 found and recalled), one more extra fact comes per ten questions, and one chit-chat
line of 70 recalls a fact ("Apa ibu kota Australia?" shares "kota", city, with a keyword "ganti kota"). The misses
left are words the keywords did not include ("перебираемся" against "переселение", "morar" against "moradia"):
keywords are as good as the agent that writes them, and facts stored without them (older ones, auto_extract,
mirror_memory_writes) are found by their words only until enrich gives them some.

What it does not do: find a fact through a synonym or another root that no keyword names ("бэкапы" for "резервные
копии", Spanish nada for natación, 十月 for 10月). It can surface a fact that shares a word in another sense:
"nightly backups" → "night market"; Hindi बनाना is both "to cook" and "to make a plan"; Portuguese dirigir is "to
drive" and "to run a company". One shared word is not enough when other facts hold it too: "What day is today?" recalls
none of the facts that happen "every day" (0.17.8), but when one fact alone says "through the week", "How's the
weather this week?" still recalls it, as "wifi password?" recalls the one fact about a password: one word cannot tell
the two apart. In a script without capitals (Hebrew, Hindi, Arabic) a name cannot be told from a common word, so
there one shared word recalls as it did. Stems are search keys, not lemmas: an irregular form can miss (ponies / pony). A
negation counts for the whole clause up to a "but" ("likes cats but not dogs" is detail, not a conflict), not for a
single word. auto_extract recognises English, Russian, Chinese and French phrasing. French covers a sentence that
opens with a preference (Je préfère…, J'aime…, J'adore…, J'utilise…, Je n'aime pas / plus…, Je n'utilise pas…) or with a decision made together (Nous avons décidé…, On a choisi…); straight and curly apostrophes.
J'ai décidé… is left out, as "I decided" is in English, and so is Nous préférons…. Only the start of a sentence
counts: Moi, je préfère le thé. and Oui, je préfère… are not extracted. Replies to the assistant are skipped
(J'aime bien ça, Je n'aime pas du tout ça, Je préfère celui-ci, On a choisi celle-là), and so is a sentence
that opens with si; a condition later in the sentence is not seen (Nous avons décidé de partir si la météo le permet. is stored, as its English equivalent is). The refusal of instruction-like text, of a
credential with its value and of a request to send one covers all eighteen languages (0.15.1; English and Russian
before). The
main sets of all eighteen languages run in CI (tests/test_holographic_languages.py).

Search by meaning (optional, online)

Off by default, and the one part of the plugin that goes online. Words, keywords and HRR stay the recall; this adds
an embedding model you choose for the questions that share no word with their fact ("Which machine copies our data
while we sleep?" against "The nightly backup runs from the OES box at 2 am."). It lives in its own module
(embeddings.py) and needs no new dependency (the standard library's HTTP client, numpy as for HRR). With it off
there are no calls and no vectors, and search scores (to six decimal places) and recall are 0.13.0's — a test
pins recall on six English facts and four queries (test_recall_is_identical_to_main); what
else 0.14.0 changes is the prompt line (see the CHANGELOG), a search_by_meaning field in stats, and the empty
fact_embeddings table a writable session creates.

plugins:
  hermes-memory-store:
    embeddings: on
    embedding_url: https://api.openai.com/v1   # any OpenAI-compatible embeddings API
    embedding_model: text-embedding-3-small
    embedding_key_env: OPENAI_API_KEY           # the NAME of a variable (e.g. in ~/.hermes/.env), never the key
Endpointembedding_urlembedding_modelKey
OpenAIhttps://api.openai.com/v1text-embedding-3-smallOPENAI_API_KEY
Geminihttps://generativelanguage.googleapis.com/v1beta/openaigemini-embedding-001GEMINI_API_KEY
A local server with an OpenAI-compatible API (Ollama, Text Embeddings Inference, Infinity)e.g. http://127.0.0.1:11434/v1the model it servesnone needed

The first two were measured (below); a local server goes through the same API and was tested against a local test
server only. Other keys: embedding_dimensions (shorter vectors — only for a model that can shorten them, like
OpenAI's text-embedding-3-*; leave it unset otherwise: a server may answer 400 to the field rather than ignore it,
as bge-m3 behind an aggregator does, and the plugin's status then says so), embedding_timeout
(3 s, at most 6: the core gives a provider's recall 8 s), embedding_min_score (an absolute floor of similarity; off),
embedding_prefetch_facts (1), embedding_weight (0.3), embedding_space (a label of your own: vectors are kept
per address, model, dimension and this label, so when a server starts serving another model under the same address
and name, change it and every vector is made again), embedding_catch_up_seconds (8, from 2 to 60: how long the
fact vectors after a turn may take; below).

A store that had embeddings from another deployment may carry a column of its own vectors in facts (e.g. a
facts.embedding BLOB). Since 0.14.3 no column the plugin did not create reaches a tool result (0.14.2 dropped
binary values only); nothing reads it either, and ALTER TABLE facts DROP COLUMN embedding (SQLite 3.35+) removes
it — unless an index or trigger of the old deployment uses the column: drop those first.

How it decides

  • Fact vectors are made after a turn, on the core's background worker, by a session that may write (the main
    agent, and a cron job unless cron_writes: false; never a subagent): requests of up to 32 facts within
    embedding_catch_up_seconds (8 s, at most 32 requests), so a store catches up within a few turns (100 facts: 4
    requests to a hosted API, about 0.3 s each) and no reply waits for it in the core's usual path. The first request to
    an endpoint carries 4 facts, and each answer's time sizes the next batch: as many facts as that speed fits into half
    of the time left — on a local server at most into half of embedding_timeout, since such a server often answers one
    request at a time and a message's vector before a turn may wait behind the batch — and one fact whenever one is
    expected to take less than the time left. Nothing is sent that would not fit, but for the turn's first request when
    it has most of the budget (a wait for the provider lock can take the rest). A batch that runs out of time is no
    failure — the server may be slow, not down: the next batch is smaller (a quarter, at the same time left); a failure
    is one fact that runs out of the turn's first request, or 3 batches in a row (0.16.3, issue #5: a 4B model on a
    local NPU server took 7.7 s for 32 facts against the 7 s left, and 0.16.2 sent those 32 again after every pause
    without ever storing a vector). With that server's numbers, 119 facts of 570 characters fill in 5 turns, about 25 a
    turn (a fake endpoint at its speed, not the server itself); a reviewer's model puts 2,000 facts at about 72 turns at
    8 s and about 10 at 60 s. Speed is measured from sending, and not from a batch a message was sent during (on a
    one-slot server it waited with it); a fixed cost per request still counts as speed, which only makes batches
    smaller. A longer embedding_catch_up_seconds fills a store sooner and delays no reply, but the core runs every
    provider's after-turn work on one worker in turn, so the next turn's extraction waits for it, and at shutdown the
    core waits 5 s for that worker: what comes back after that is not stored (the store is checked before anything is
    written), though the request itself is not called back. fact_store stats shows the budget and what the endpoint's last batch did
    (after_turn; the endpoint is shared by every store a provider opens). A vector is stored in its own
    table, fact_embeddings, with the model, the address (a query string by a short hash of it) and a hash of the
    text it was made from: a changed text, another model, address or embedding_space gets a new one, and forget, end and remove delete it with the fact.
  • Before a turn one request embeds the message — unless it is a greeting or a thank-you, has fewer than two words
    that carry meaning (five characters of Chinese, Japanese, Korean or Thai), looks like it holds a secret (below), or
    no fact has a vector yet. Meaning then reorders the facts that share words with the message (embedding_weight of the score),
    and may add one fact that shares none (embedding_prefetch_facts), after the words' facts and marked
    (by meaning), when it stands far out: its similarity is at least sqrt(2 ln N) + 1 standard deviations above this
    message's mean over the N facts with a vector (the closest of N unrelated facts reaches about sqrt(2 ln N) by
    chance, so the bar rises with the store; it needs 20 vectors, below that only embedding_min_score admits a fact),
    and it mentions whoever the message names — by a heuristic: every capitalized word inside a sentence and every
    known entity the message mentions. In meaning "When is Peter's birthday?" is as close to "Lena's birthday is on
    June 3" as a paraphrase is to its fact; only the name tells them apart. The heuristic is off in German (every noun
    is capitalized), sees no names in Chinese, Japanese, Korean or Thai, nor a name that opens a sentence or is written
    in lower case, and turns away a paraphrase that names a place its fact words differently ("Japan" for a fact about
    Tokyo).
  • The bar is a heuristic too, not a guarantee. Similarities in one store are not independent: when many facts are
    about as close to a message as each other (a cluster — twenty facts about one child), the spread grows and fewer of
    them clear the bar. Measured: 20 more facts about two people of the set cost 6 of the 52 paraphrases that
    cleared the bar on text-embedding-3-small. A median-based spread let more paraphrases through but also more noise (13
    of the 45 questions that must recall nothing got past the bar alone, against 8), so the mean and standard
    deviation stay.
  • fact_store search adds the closest facts by meaning after every word match when there are fewer than limit,
    with "match": "meaning" and "similarity" (bar sqrt(2 ln N) + 0.5: the agent asked, and judges). And
    even when the words fill the limit (0.14.4), it shows those of the 3 facts closest in meaning that are closer than
    the best fact the words found by more than 0.06 of cosine, stand at least 1 standard deviation above the store's
    mean for the query, and mention whoever the query names (the name check recall uses): words can fill the list
    while the fact asked for is the store's closest in meaning.
    Measured on a family store (58 facts): questions sharing no word with their fact but a name, 13 → 18 of 27; no
    word at all, 3 → 6 of 12; questions that must find nothing still got a fact by meaning in 1 of 30. Taking the 3
    closest by rank alone found as many but answered 20 of those 30 with one. Recall before a turn does not do this.
  • Any failure — network, HTTP status, a timeout (embedding_timeout for the whole request before a turn; fact
    vectors after a turn share embedding_catch_up_seconds with every wait for the provider lock, as no reply waits
    for them, and count as above;
    either way the request, decoding included, runs on a thread of its own and the caller stops waiting then, however
    slowly the server sends — and the socket is shut down when the wait ends, so the thread's read ends with the
    caller, not past it — with one interval kept honest: at the TLS wrap the raw socket is detached and the wrapped one not
    yet published, so a cut reaches neither there and the handshake runs to its own socket timeout; the retained cancellation
    then refuses the connection immediately (bounded, not interrupted mid-handshake); resolving the name has no socket yet, so
    the OS resolver bounds that, not these timeouts),
    a malformed answer, an HTTP status that comes after the caller stopped waiting — falls
    back to words without a word to the user; the endpoint is then left alone for 60 s, doubling up to 15 min, with one
    warning in the log. One request moves the pause one step at most, however many ways it fails; an answer to a
    request sent before the latest failure does not end the pause. While a request given up on is still running no new
    one starts; the next one waits for it up to half of its own time first (the answer to a batch cut short often comes
    a moment later), so a message's network waits share one embedding_timeout. A message's request that runs out of
    time while a fact batch of ours held the endpoint for part of its wait (the batch was running, or ended after a
    quarter of that wait) is no failure: recall uses the words that turn, and the endpoint is not paused. A batch of
    ours counts as holding the endpoint for twice embedding_timeout after it was given up on; a request still running
    after that is a stalled server, and fails (the status says so from then on). A redirect is a
    failure too (it would carry the key elsewhere). fact_store stats shows search_by_meaning: the status, the model,
    and how many facts have a vector, wait for one, or are never sent.

What leaves the machine. To the endpoint you name: the text of every fact (once, again when it changes) and of
every message and fact_store search query that passes the checks above. Fact texts go out after a cron job's turn
too while cron_writes is on: a job that may write also makes missing fact vectors. The key is read from the
variable embedding_key_env names — OPENAI_API_KEY when it names none — and sent as a Bearer token in the
Authorization header to whatever embedding_url is configured, so point the URL only at a provider you mean to give
that key to (a key with a space or a line break inside is refused, and errors are logged as categories, never with
the request in them).
Not sent: a fact recall would not inject (instruction-like text, markup, an image — the same check recall makes), and
a fact or a message that looks like it holds a secret — a credential word with its value, an environment assignment
such as OPENAI_API_KEY=… (not a variable named for a file that holds a file name: SSL_KEY_FILE=server.key,
API_KEY_PATH=keys/prod.json; 0.17.2), an Authorization header, the prefixes of common API keys and tokens (sk-, ghp_,
AKIA, ASIA, AIza, xox…-, glpat-), a Telegram bot token, a JWT, a PEM private key, a long encoded blob; of a message, only the first 2,000
characters are checked — and sent. That check is by pattern, so best effort: a secret in a shape
it does not know goes out with the message. forget deletes the vector, but the provider has seen the text: what it
keeps is up to its terms, and some free tiers may use what they receive to improve their models. The plugin cannot
tell whether a vector belongs to the text it sent: an endpoint or aggregator that mixes up concurrent answers gives a
search another text's closeness, silently (a user saw one fact by meaning carry the cosine of a parallel query,
issue #3; the plugin's own side is covered by a test). A local server keeps
everything on the machine; plain http is accepted only to localhost, a private or link-local address, a host name
without a dot or one ending in .local, .lan, .internal or .home.arpa (judged by the name; the address it
resolves to is not checked), and needs no key. A redirect is refused. With scope: user, chat or thread every
database has its own vectors. embeddings: off again: the next writable session deletes every stored vector.

Cost. One request per turn for the message, one per 32 facts written; at OpenAI's price for
text-embedding-3-small ($0.02 per million tokens) a thousand facts cost well under a cent. One request adds about
0.2 s (OpenAI) to 0.3 s (Gemini) before the reply (median of 12 requests from a VPS). Vectors are kept in
memory: 4 bytes per dimension per fact, 6 KB at 1536 dimensions (60 MB at 10,000 facts); embedding_dimensions
shrinks them.

Measured (2026-10-14; synthetic, nothing real: 100 facts in English, Russian, Chinese, German and Spanish; 67
questions that share no word with their fact but a name; 45 that must recall nothing; default settings):

Words only (off)+ meaning: text-embedding-3-small+ meaning: gemini-embedding-001
Paraphrases recalled before the turn22 / 6754 / 6759 / 67
A wrong fact recalled by meaning—02
Chit-chat that recalls a fact by meaning—0 / 130 / 13
Off-topic questions that do—1 / 222 / 22
Questions about someone or something not stored—2 / 102 / 10
Facts recalled on meaning alone that were the right ones—29 of 32 (91 %)35 of 41 (85 %)

The off-topic ones sit next to a stored topic ("Сделай план тренировки на неделю" → "Антон занимается в спортзале
три раза в неделю"). The others are names the rule cannot see — Chinese has no capitals (小王 / 小李), German
capitalizes every noun, so the rule is off there — and a related thing in another language ("Какой пароль от роутера?"
→ "Der Router steht im Flur."). With 50 facts text-embedding-3-small recalls 25 of 35 against 12 by words. Keywords
(above) stay worth having: they are local, free, and work with this off.

On a real store (2026-10-14, 0.14.1): a family assistant's 58 facts, mostly Russian, each with keywords from
hermes-dreaming's nightly job; a copy of the store, questions
written by a model from the facts, only numbers kept, and a run counted only if no request to the endpoint failed.
Natural questions: 57 of 58 recalled, with words alone and with meaning alike. Questions sharing no word with their
fact but the name of the person they ask about: 7 of 27 by words, 8 with meaning. Questions sharing no word at all:
0 of 12 either way. No wrong fact, and nothing from chit-chat, off-topic questions or questions about people not
stored (0 of 30), came in by meaning. On this store keywords and names do most of the work and the default bar added
one recall in these questions; a lower bar on the same questions: margin 0.5 added two with nothing wrong, 0 added
four but brought two wrong facts and three of the 30 other questions. The synthetic table above, whose facts carry no
keywords, is a separate benchmark; one store does not show what another will do.

Rescoring by a cross-encoder (optional, online)

Experimental — it does not raise hits. The gate agreed on in
#9 did not pass on either set
(branch A needs +5 points and there are none; branch B allows no lost target and each run loses at
least one — the embeddings-on hold-out loses two),
so the option ships off by default and states its trade plainly. Three independent runs, all with
jina-reranker-v2-base-multilingual Q4_K_M, rerank_scale: logit, window 0.05, weight 0 — the
maintainer's held-out set twice, with embeddings off and on (bge-m3 Q4_K_M vectors from
gpustack/bge-m3-GGUF, sha256 6d39681b…a82fc06, llama.cpp b11836, CPU — all 199 facts
vectorized per language):

runhit@5 off → onnon-target in top 5lines/turntargets lost / gainedconfused rows
this PR's test split (90 queries)43 → 43108 → 66 (−39 %)1.98 → 1.271 / 06 → 4
the maintainer's hold-out (126 queries)64 → 64249 → 154 (−38 %)2.24 → 1.441 / 124 → 18
the maintainer's hold-out, embeddings on (126 queries)81 → 81251 → 150 (−40 %)2.38 → 1.532 / 124 → 18

lines/turn in all three rows is per real question. The hold-out figures 1.84 → 1.21 and 1.97 → 1.30
quoted earlier were an average over all 195 rows, chit-chat included (denominator pointed out by the
maintainer, 2026-10-24); per question: 2.24 → 1.44 and 2.38 → 1.53.

With it on, a third fewer lines reach the model and ~40 % fewer non-target facts, with fewer
confusions in all runs and noise (chit-chat, no-answer) unchanged; the cost is no gain in hits and
about one target per hundred questions — one in the 126-question emb-off run, two in the emb-on rerun —
lost on questions holding several facts of one kind (q055, h42, h34). With it off, nothing changes — the off rows match main (195 of 195 on
the hold-out set), and failed rerank calls were 0 in all three runs.

Off by default (rerank: off): words, keywords, HRR and, when on, meaning decide as above, with 0.17.8's scores and
recall — a test pins them on six English facts and four queries (test_recall_is_identical_to_main,
golden blocks per mode). Turn it on to add a second stage: a
cross-encoder reranker (Cohere's /v1/rerank shape; OpenAI has no rerank API) reorders the candidate pool before a
turn, after the blend. It lives in its own module (rerank.py), needs no new dependency, and makes one synchronous
HTTP call at recall time; a failure or a slow answer only keeps the current order and pauses the endpoint (60 s
doubling to 900 s). One mode since the 2026-10-22 review: the two designs were compared in
#9 and collapsed into one —
rerank_mode and rerank_margin are gone (a config that still carries them is ignored with a warning):

plugins:
  hermes-memory-store:
    rerank: on
    rerank_window: 0.05              # δ: cut the tail below this distance of the best score; 0 = keep all
    rerank_url: http://127.0.0.1:8080/v1/rerank   # llama-server --reranking --embedding --pooling rank
    rerank_model: local-reranker       # the endpoint's --alias

The cross-encoder reorders the pool; membership stays with coverage (the words' own criterion, exactly as in
off), so the reranker never adds a fact the words would not admit. It can still lose one they did admit: with
rerank_window > 0 the tail below δ of the best score is cut before the top five is taken by rank — if the
cross-encoder is confidently wrong at the top, a fact the words admitted disappears instead of slipping places.
The cut never drops below two scored candidates — that minimum is built into the window itself (review of
2026-10-23), so a confidently wrong top cannot squeeze recall down to a single fact. Meaning- and entity-based
facts compete in the same rank as the lexical pool. A failed call gives exactly the recall of rerank: off.

Which servers work: it is tested against llama-server. Any server that answers the Cohere shape
({"results": [{"index", "relevance_score"}]}) without a key should work too. The client sends no key, so a
cloud API that requires one (Cohere, Jina) refuses every call, and Text Embeddings Inference's own /rerank
answers a bare list, which is refused as malformed. Either way the endpoint is paused and recall stays as with
rerank off.

Other keys: rerank_top_k (40: how deep a pool the cross-encoder sees; the first stage's limit is widened to
match), rerank_weight (0 = order-only, recommended; >0 = the min-max-normalised cross-encoder score blended into
the blend score; the window applies to order-only runs), rerank_timeout (3 s — additionally capped so that embedding_timeout + rerank_timeout stays under
the 8 s the core gives a provider's prefetch: with embedding_timeout at its maximum 6 s the rerank
timeout clamps to 1.5 s; when the wait ends the request's socket is shut down too, so a worker cannot read past
its caller's deadline — except the TLS handshake, whose socket sits between the two publications and so runs to
its socket timeout, the retained cut refusing the connection right after; resolving the name has no socket yet,
the resolver bounds that) and rerank_scale —
the endpoint's score scale, set explicitly: logit (raw llama-server logits; the default, no probing) or
unit ([0,1], Cohere-style). Set unit for an endpoint whose scores are already in [0, 1]: on the
default logit the sigmoid runs on scores that were normalised once already, the gaps shrink and the window
keeps more than it should. A value outside that pair is a config problem that names the key and turns
memory off for the session, like any other out-of-set value — never a silent fallback.
A query that holds a secret is never sent. A message over 500 characters (a night job's prompt, a pasted log) is not rescored either:
with a fact beside it, it could run past the model's 1,024 tokens, and the server refuses the whole call; its order
stays the rerank-off order, and it is not a failure, also while the endpoint is paused (skipped_long). Each fact is
cut to what the message leaves of 1,000 characters for the pair; with XLM-RoBERTa's tokenizer (jina-reranker-v2,
bge-reranker-v2-m3) a token covers at least one character, so the pair stays under the limit for CJK and emoji too.
With rerank_weight: 0 the cross-encoder judges text only and ignores trust: a fact whose trust
was lowered (by feedback or by a dreaming plugin) can move back to the top of what is admitted.

A call asks no more than the endpoint answers in half of rerank_timeout. A reranker on a CPU scores the pairs at a
steady speed, so its time grows with every fact it is sent: jina-reranker-v2 Q4_K_M on 2 threads of a 4-core VPS
answered 5 short facts in 0.4 s, 15 in 1.2 s and 40 — the default rerank_top_k — in 3.3 s, over the default
timeout, so every large pool failed and paused the endpoint. Each answer now tells the plugin the endpoint's speed:
seconds per character of the query-document pairs, counted from the moment the connection was ready. A call that
timed out after connecting tells it too, as a lower bound; one that timed out before (DNS, TCP, TLS, a proxy) tells
nothing. A call sends as many of the first candidates as the slowest of the last five speeds fits, in characters, in
half the timeout — an estimate, so a call can still time out — and never fewer than two. With rerank_weight: 0
the candidates not sent keep their order behind the scored ones, as the tail beyond rerank_top_k does; with a
weight above 0 they enter the blend with the lowest rerank score, as that tail does. A fast endpoint is sent the whole
pool. Before the first answer the speed is unknown and rerank_top_k goes out: on a slow endpoint the first large
pool can time out once, and the next call then sends about half of its characters.

Measured on a real llama-server with bge-m3-reranker-v2 (Bge-M3-568M-Q4_K_M), three languages (en/ru/zh),
60 questions and 15 that must recall nothing (tools/benchmark_rerank.py --endpoint=…). The first table is
the PR author's CPU box; the same run on a GTX 1050 Ti (2026-10-22) gives p50 32–35 ms / p95 47–51 ms for the
four rerank modes and p50 4 ms off — the same order of cost, and recall/wrong/noise identical:

moderecall@5wrong-entitynoisep50 (ms)p95 (ms)
off100% (60/60)00%48
window off, weight 0100% (60/60)00%3852
window off, weight 0.5100% (60/60)00%4052
window 0.05, weight 0100% (60/60)00%3851
window 0.05, weight 0.5100% (60/60)00%3751

All five rows were measured on the earlier design, before the 2026-10-22 collapse into one mode
(reorder_before_gate = window off, gate_reads_score = window 0.05) — they are not the current result:
the head-commit table is in results/BENCHMARK.md, and the labels in it keep the historical names.

This store's recall was already 100% before the turn (its data are calibrated to ≥95%): the reranker keeps it there.
The honest number this table adds is the per-turn cost — 3-4 ms to 40 ms median, ~52 ms p95 on the author's box,
32–35 ms median / 47–51 ms p95 on a GTX 1050 Ti — the price the maintainer's acceptance asked to see next to the
gain. On CPU (results/bench_cpu_t4.jsonl, the head): llama-server with the same model, capped at
threads 4 on a Ryzen 7 2700 — 176–184 ms median / 393–413 ms p95 per call, 0 fallbacks. The earlier
CPU numbers (94–101 ms median / 212–229 ms p95) were measured on a different processor — a
Ryzen 7 5800H laptop, 7f1efd6, before the words-gate fix (results/bench_p8_cpu.jsonl): same model
and harness, not comparable across machines — see docs/contrast-benchmark.md. A second reranker of
different quality and a larger public set are run before merging (issue #9).

Choosing a reranker model: jina-reranker-v2-base-multilingual Q4_K_M is the CPU recommendation — p50 121 /
p95 152 ms at 4 threads (620 calls per model, results/bench_cpu_gate_jina.jsonl) — and it is
CC-BY-NC-4.0: non-commercial use only (both upstream jinaai and the gpustack GGUF). For work that
must stay commercial, use bge-reranker-v2-m3 (Q4_K_M, Apache-2.0): the same CPU run gives p50 343 /
p95 448 ms at 4 threads — it misses the 300 ms bar on CPU and stays the GPU default, but its licence is safe
for commercial use. The plugin ships neither model.

Data and privacy

  • Facts live in one SQLite file (db_path), in the Hermes home of the profile, readable by its owner only (POSIX
    0600). Nothing is sent anywhere unless embeddings or rerank is on (see Search by
    meaning
    and Rescoring by a
    cross-encoder
    for what is sent, and to whom). A db_path outside
    the Hermes home is listed for hermes backup.
  • Subagents and flush contexts read the store but never write to it (facts, feedback, counters, extraction). A cron
    job writes facts only through its explicit tool calls (cron_writes, on by default); the one automatic write it
    makes is a fact's vector when search by meaning is on.
  • With auto_extract on, user sentences matching the preference/decision patterns are stored after each turn and,
    outside scope: user, before context compression and at session end.
  • retrieval_count counts how often a fact was handed to the model; nothing else is tracked.
  • For statistics, logger.debug stage traces name fact ids, never their text: embed — the fact vectors stored
    after a turn, the facts the meaning stage admitted before a turn and the meaning matches a tool search returned;
    rescore — the pool and the candidates the cross-encoder would score and its #id=score answer, or why no
    made. Set logging.level: DEBUG in config.yaml to see them in logs/agent.log; at the normal level they are
    not emitted, and there is no extra switch for them.
  • Each version keeps its source: platform, chat type, chat / thread / session id, the author of the turn (id and
    display name) and, from 0.15.0, its origin (said / confirmed / inferred). Chat names are not kept. Each confirmation
    keeps its time, session id and the person's id, no text. forget deletes the source and the confirmations with
    the fact.
  • merge (0.19.0) removes a duplicate from the present but keeps its text, versions, sources and confirmations in
    the file under its own id; to erase it, forget the copy (or the keep, which takes its copies with it).
  • forget deletes a fact for good (see History); copies outside this database
    (Hermes's session transcripts, backups, other tools' caches) are not the plugin's to rewrite.
  • Dependencies: numpy (HRR algebra) and snowballstemmer (word forms; pure Python; Polish stems need 3.1 or
    newer). Without numpy the provider
    works with keyword search only and says so in its prompt block; without snowballstemmer words compare exactly.

Database and downgrades

Schema changes are additive: 0.4.0 adds a meta table (schema version, HRR dimension), an entity_aliases table and
an entities.name_key column; nothing older is dropped or rewritten, and the new column is nullable, so earlier
versions keep reading and writing the file. When this version opens a
file an older one wrote to, it reconciles: missing keys are filled in, entities with the same key are merged (links,
aliases and names kept), aliases of entities the older version deleted are dropped, and vectors whose size does not
match hrr_dim are re-encoded. Only a writable agent does this; subagents, cron and flush contexts never migrate.

0.8.0 adds fact_versions (with its own FTS index) and forgotten; facts keeps its columns, so older versions
and tools that read facts directly keep working, and see only current facts. A file an older version wrote to is
reconciled on open: a fact without a version gets one (start unknown, source unknown), a text changed by an older
update becomes a new version with the old one closed, and a fact an older remove deleted is forgotten (its history
deleted, a tombstone written). See docs/history.md.

What is tested: this release, then 0.1.1, then this release on one database in CI (adds, edits, removals, aliases, a
keyless duplicate entity written by 0.1.1), plus the reconciliation cases above. What a downgrade does not keep:
aliases made after 0.4.0 are visible to 0.1.1 only through the legacy entities.aliases column; and with
scope: user, chat or thread, an older version knows nothing of per-user or per-chat files (0.6.x and older do not
know chat / thread at all and fall back to profile), so it reads and writes the shared db_path again. Those
files stay untouched (no leak), but new facts would land in the shared store.

0.14.0 adds fact_embeddings (search by meaning): created by every writer, empty unless embeddings is on, never
read by older versions. A fact an older version deletes leaves its vector until this version next opens the file for
writing (a subagent or another read-only context does not drop it); a text an older version changes no longer matches its vector's hash and is never compared until a
new vector is made.

0.16.0 adds entity_aliases.source (with the default legacy: an alias written before, or by an older version
since, is of unknown origin) and entity_tombstones, both created by a writer, and records the extraction rules in
meta.entity_rules. A context that does not migrate still reads names by their keys and aliases once any writer of
0.4.0 or later has migrated the file. An older
version reads the file as before; one that also writes to it may create a name delete_entity removed again from a
new fact's text, and this version's next open does not notice: run rebuild_entities after going back and forth.

0.17.0 changes no schema. An older version reads and writes the file as before, but does not keep the rule of 0.17.0:
it resolves a declared name through an alias a fact's text gave, as 0.16.x did. After going back and forth, read
review (alias_candidates) and entities (text_aliases). 0.17.1 changes no schema either.

0.19.0 is the first release an older one cannot write after. It adds fact_merges and the version status
merged. A file without a merge stays at history schema 1 and every release since 0.15.0 keeps reading and writing
it. The first merge raises meta.history_schema to 2; from then on a writer of 0.15–0.18 refuses the file
(HistoryTooNew) and does not stamp it back, so going back below 0.19.0 after a merge switches the main agent's
memory off
until 0.19.0 is installed again or the file is restored from a backup made before the merge. Setting the
schema back by hand is no fix: an older release's forget of a keep would leave its copies' text in the file. An
older writer already open when the schema rose is not stopped; what it leaves (copies of a keep it forgot, a merge
row of a copy it forgot) this version cleans on its next writer open, a tombstone for each erased copy. Readers of
0.15–0.18 (subagents, flush, cron without cron_writes) see only the keep; their as_of shows a copy under its own
id without the merge-day clip. Releases before 0.15.0 do not check the schema: they read facts as before, and a
copy's text they add again is a second fact, not folded back. The schema mark is per file, so with scope: user,
chat or thread a merge in one file does not lock the others. See
docs/history.md.

Development

Tests import Hermes core from a checkout, as an installed plugin does:

HERMES_AGENT_REPO=~/.hermes/hermes-agent PYTHONPATH=~/.hermes/hermes-agent python -m pytest -q

CI runs the suite on Linux, macOS and Windows against hermes-agent main, ruff check . with the rules in
pyproject.toml, and hermes plugins validate, the check the plugin catalog runs at admission. 0.11.0: 612 tests pass without the core (2 skipped; the auto-extraction tests
need it), 600 with snowballstemmer 2.2.0 (14 skipped: no Polish stemmer).

The RECALL_MARGIN gate runs on recorded rows, never on a live model: tools/provenance_probe.py --modes off --recall-margin <v> per margin writes results/provenance_*_m<v>_<emb>.jsonl, and
tools/mcnemar_gate.py --margin-pair <files> pairs the two runs per (set, lang, qid) and reports the four
guards — no target lost, noise not higher, confusion not higher, pooled exact McNemar p < 0.05. The pair
needs rows of a single mode: the probe writes three by default (off,reorder_only,gate_only) and the
gate refuses mixed runs, so pass --modes off. One endpoint, one git_sha, one scale and one emb
per comparison, and inside a pair only the margin may differ — a pair whose query, kind, split or targets
disagree is refused instead of reported.

Original plugin by dusterbloom (hermes-agent PR #2351); its history is in hermes-agent
(git log -- plugins/memory/holographic). MIT licence. The bundled language data live in dictionaries/ (the list
with sources in its __init__.py, rebuilt by the scripts in tools/); the Traditional → Simplified Chinese table in cjk_fold.py is
derived from OpenCC under Apache-2.0 (LICENSES/OpenCC-Apache-2.0.txt), the Russian
derivation pairs in ru_derivation.tsv.gz from DerivBase.Ru (UDer 1.1) under Apache-2.0
(LICENSES/DerivBase.Ru-Apache-2.0.txt), the Japanese spelling variants in
ja_variants.tsv.gz from SudachiDict's synonym source (Works Applications) under Apache-2.0
(LICENSES/SudachiDict-Apache-2.0.txt), the Chinese words in zh_words.txt.gz
from jieba's dictionary (Sun Junyi) under MIT (LICENSES/jieba-MIT.txt), the English
derivation pairs in en_derivation.tsv.gz from Open English Wordnet 2025 (the Open English Wordnet team) under CC BY
4.0, built on Princeton WordNet 3.1 (LICENSES/OEWN-CC-BY-4.0.md,
LICENSES/OEWN-WNDB-License.txt), and the German compounds and German and Spanish
verb forms in de_compounds.tsv.gz and de_es_forms.tsv.gz from Wikidata Lexemes under CC0
(LICENSES/Wikidata-CC0.txt).

README from the reviewed commit e9c47ea; it updates when the author re-pins.

More by itpartypattaya

4 plugins by itpartypattaya in the catalog.

itpartypattayahermes-dreamingNightly deterministic memory consolidation ("dreaming"): corroborates candidate facts with what humans actually said on distinct days, proposes the few worth keeping with a verbatim edit anchor, asks whether stale memory entries still hold, and wakes the agent only when there is work — including a memory file close to its char limit. Fact sources: the holographic store, a SQLite or JSONL export, or none. Secrets and instructions are screened by the dream's own rules plus a verbatim copy of the holographic plugin's screen (scripts/holo_screen.py, MIT, same maintainer), applied to stored facts and human messages before anything reaches memory, the report or the model. Portable Agent Plugins v1 package (plugin.json + skills/); no tools, hooks, middleware or env vars. Disclosure — by default the scripts make no network or LLM calls and open state.db and every fact store read-only (SQLite mode=ro; with holographic >= 0.15.0 including its fact_versions/fact_confirmations tables, to say who stands behind a fact); they also read Hermes config.yaml's memory/timezone keys and the holographic plugin's db_path/scope/cron_writes/min_trust_threshold (the floor recall applies, so a rejected fact is only reported when the store still trusts it) and, to say so in install.py --check, embeddings/mirror_memory_writes, the pending/memory approval queue, cron/jobs.json (to find their own jobs) and — only when chat_stores is set, with holographic scope user/chat/thread — the per-chat databases next to the store (counts of every one; facts of the chats listed in promote_from). An optional cloud judge (judge in dreaming.json, off by default; off, its code is not even loaded) may, once the owner enables it, send for each mention the rules counted the fact's text and that person's own words (no reply quote, link, id, sender or time; names inside the text may be included; screened for secrets and instructions first), only from the chats, sources and chat stores the owner lists, over HTTPS to the providers the owner lists, in the owner's order: OpenRouter's Decisions API (model typesafe/jev-1.14, Jev by TypeSafe; the data reaches OpenRouter and TypeSafe) and/or OpenAI's Decisions API (model gpt-6-luna); when the first gives no usable answer the same pair may reach the next. A verdict can only remove a counted mention — from promotion, confirmation, fact decay, "said again" after a rejection, the selection of new facts and conflicts, and evidence. Keys OPENROUTER_API_KEY / OPENAI_API_KEY are read from the environment or, since Hermes withholds these provider keys from cron scripts, straight from their own line in $HERMES_HOME/.env; the judge's calls are billed to those keys. It keeps a 90-day verdict cache and a 90-day local journal of the pairs sent in $HERMES_HOME/cache (the journal is deleted on nights the judge is off); the nightly spending cap ($0.05 by default) is soft; data already sent cannot be recalled and falls under the providers' retention terms. The user-run installer copies three gate scripts into $HERMES_HOME/scripts, seeds $HERMES_HOME/dreaming.json and, through Hermes' cron Python API, creates (or with --rebind re-points) two agent cron jobs bound to this plugin's own skill with only the memory toolset, plus one no_agent job (every 15 min, no model) that reads the fact stores to keep a hash index of diary lines and, for facts holographic >= 0.8.0 has forgotten, rewrites the dream's own cache files, judge files and diary without them, deletes the output files of the dream's own cron jobs that mention them (by id or verbatim text) and deletes the set-aside copies of its own broken state files; with a chat --deliver target the reports are attached as a session you can reply to. The agent runs send recent human messages (up to 200 / 40k chars, each with its chat type and sender id) and fact snippets to your configured model; state goes to $HERMES_HOME/cache and a diary to $HERMES_HOME/memories/DREAMS.md (configurable); memory files are changed by the agent only through the regular memory tool, and with the holographic provider it also adds candidate facts via fact_store (with holographic >= 0.15.0 marked said or inferred; with holographic >= 0.8.0 also change / end / correct / forget, the last three only on a human's own answer; forget is irreversible), lowers trust on human-rejected facts via fact_feedback, with holographic >= 0.15.0 raises trust (fact_feedback confirmed, +0.05, at most 10 facts a night) on facts people repeated on two or more different days and, when keywords.enabled is set (off by default), adds search keywords to facts via fact_store update. Since 3.4.0, with holographic >=0.19.0 and profile scope, it also reads fact_merges and current version ids to propose disjoint duplicate pairs. It saves immutable proposal numbers, both full texts, version ids, origins, keywords and the provider database path in cache/dream-proposals.json; open proposals expire after 180 days and closed ones are removed after 30 days. Full pair texts are included in the agent report and therefore reach the configured agent model. Old facts are merged by fact_store only after the adult memory owner approves the exact proposal; the response checks the database and uses expect to refuse changed versions. Scheduled extraction may merge only a newly added paraphrase from its own session into an existing fact, never old facts from other sessions. The scripts themselves still open the fact store read-only. The sweep also cleans the state of merged copies and archives their hashes so a later forget can remove older untagged diary copies; merging itself does not erase the diary. No new credentials, network providers or tools are introduced. Version 3.4.2 clarifies that forget cleans the store and the skill's own files; Git history, external database dumps, system backups and Hermes session history require separate cleanup. Reports distinguish a successful store deletion from a retained external copy and do not promise that retrying removes a documented history boundary. Version 3.5.0 adds explicit cron maintenance: known stock prompts are synchronized while custom prompts are preserved; --update-existing applies only explicitly supplied settings and --job-id scopes the entire operation. cron.expected_roles controls desired creation without pausing existing jobs. --upgrade validates the request before copying gate scripts and updating cron through public APIs, then checks the result. Checks expose paused jobs, execution modes, missing roles and duplicates; no automatic activation or job deletion. --inherit-model explicitly clears both model and provider pins. The public cron update API has no CAS; avoid concurrent configuration edits during installation. Version 3.6.0 adds opt-in read-only task-outcome summaries and recorded memory-attempt explanations from holographic >=0.20.0 (experience_schema=1). It reads bounded attempt/version references and agent-reported outcomes; usefulness does not change truth, trust, ranking or MEMORY.md. It makes no new model or network calls and keeps no second correction-text cache. Explicit --memory-store must be in chat_stores.promote_from; stores are never joined. task_outcomes: true adds only summary counts to the ordinary nightly result and creates no extra wakeup. Missing capability is reported as unsupported, not as a successful empty journal. Human corrections reported by the agent are not authenticated human confirmation. Version 3.6.1 reuses parsed numbers within a bounded in-memory nightly cache, reducing repeated work without changing contradiction rules. Version 3.6.2 copies the cron gate scripts atomically (an interrupted install keeps the previous gate) and names any cron job the installer refuses to change with the fix. Version 3.7.0 keeps the whole-store duplicate search within its nightly budget on large stores (it also narrows candidates by the numbers and names the pair rule requires; the rule still decides every pair) and names the store in task-outcome summaries. Version 3.7.1 derives each message's data once per nightly pass, with memory bounded by corroboration.memory_mb (default 256 MB; above it data is rebuilt on demand, never changing decisions). Version 3.7.2 makes the quarantine screen about 40× faster on long messages with byte-identical decisions (re-vendored holographic screen 0.22.1). Version 3.7.3 fixes an install-time scanner false positive in that screen's start list (3.7.2 was refused by hermes plugins install) and runs hermes plugins validate in CI. Version 3.7.4 changes no behaviour: the vendored holographic screen is copied whole and checked before it is written. Version 3.8.0 no longer promotes an agent's conclusion nobody confirmed (holographic origin inferred) to MEMORY.md by any corroboration route; people repeating it offer it for confirmation instead, and stores without origins are promoted as before; --explain says why, and --memory-explain keeps the provider's count of candidates that changed before the record. Version 3.8.4 keeps, for each memory entry, hashed word stems and numbers plus a 60-character preview (none for an entry with a secret) in the loss-guard snapshot in $HERMES_HOME/cache, so an entry the agent compressed, reworded or merged is no longer reported as lost; the alert names lost entries by that preview. Version 3.8.5 has the nightly agent change memory entries with the memory tool's patch (Hermes 0.22 removed replace; replace remains only as the fallback for Hermes 0.21), counts a patch staged in the pending/memory approval queue as waiting, and compares a lost entry only with entries the night wrote (also across a split, with numbers compared as numbers); for an entry of fewer than two word stems the snapshot also keeps hashed stems of its short words or, for one with no words and no secret, hashed character trigrams; state files with a UTF-8 BOM are read. Version 3.8.6: install.py --check counts a judge key only from its own line in $HERMES_HOME/.env, the one a cron script can read, and names DREAM_* variables set only in the terminal it runs in, compared with their lines in that .env (a DREAM_* value may be printed, a key never is); it and a new read-only dream.py --nightly-status read the dreaming jobs' next_run_at in cron/jobs.json to say when Hermes will not fire them (a gateway that is down, a job paused or with no next run). The nightly prompt never patches through another entry's whole text and keeps an absorbed entry when a merge fails; the loss-guard snapshot also keeps hashed full dates and each entry's hashed head key. No new credentials, network calls, tools or writes.Communitymemory27itpartypattayahermes-cronOperating skill for Hermes cron: lifecycle rules (pause vs disable vs remove, in 30m vs 30m), four ways to avoid needless LLM runs (no_agent, wakeAgent pre-check, monitor, [SILENT]), verify-after-change, a layered runbook and a read-only doctor that complements hermes cron doctor and reports one error across several jobs as one scheduler fault and a cron store that cannot be written. Portable Agent Plugins v1 package (plugin.json + skills/); no tools, hooks, middleware or env vars. Disclosure — skill only, no network; the bundled doctor reads ~/.hermes/cron/jobs.json, ticker stamps and executions.db (or a named profile's store with --profile) read-only, checks the cron directory's permissions and free space without writing, and runs the local hermes cron doctor (argument list, no shell), whose job loader may repair malformed jobs.json entries (--no-builtin skips it).Communityautomation6itpartypattayahermes-field-notesField notes for a Hermes install: search known pitfalls before debugging and record new ones with root cause; register every local core edit as a patch note with a declarative checks file, then verify read-only after each hermes update (OK / MISSING / UPSTREAMED / N/A / UNKNOWN with reason) and list core files changed outside the registry. Ships 42 pitfalls and 10 patch recipes checked against 0.22.0. Optional pinned Telegram dashboard edited in place by an hourly no_agent cron job. Portable Agent Plugins v1 package (plugin.json + skills/); no tools, hooks, middleware or env vars; never edits the core or runs patch scripts. Disclosure — its scripts only read the Hermes core (pyproject.toml, read-only git status/diff) and never edit it or run patch scripts; the skill does walk the agent through hand-editing core files from its patch recipes and re-running the owner's re-apply scripts, each only after the user says yes. Network: Telegram Bot API only when the dashboard is enabled (reuses the agent's TELEGRAM_BOT_TOKEN from env or ~/.hermes/.env), and the public GitHub API without credentials (latest release plus compare against the installed commit SHA, cached 12 h; upstream.check=false disables). Writes: the notes store (~/.hermes/field-notes), ~/.hermes/field-notes.json, state and lock files in ~/.hermes/cache, a copy of the cron script in ~/.hermes/scripts when install.py runs (the skill runs it on first use), and one hourly no_agent cron job only when the user runs install_cron.py (it reads ~/.hermes/installs/*/facts.json to find Hermes' own Python and re-runs itself with it). Without a dashboard chat, install_cron.py sends the job's alerts to the Telegram home channel when TELEGRAM_HOME_CHANNEL is set (read from env or ~/.hermes/.env), else local.Communitytools2

Developer Mode

Explore built-in Hermes toolsets and actions.

HermesAgent

Developer?

Create a Hermes extension for Desktop or agent workflows.

Developer Section