All plugins

gbrain-plugin

Community (non-official) plugin wiring GBrain (garrytan/gbrain, MIT) into Hermes as a local memory layer over MCP: /gbrain status command, desktop starter, and per-profile Postgres setup/update/maintenance scripts. Not affiliated with upstream. Disclosure — the plugin itself only adds a /gbrain status command and a Desktop pane; GBrain is installed by the user via scripts/setup.sh, which installs bun (curl|bash), an unpinned gbrain from GitHub HEAD, and provisions PostgreSQL with sudo.

kerrz20205Memoryv1.2.1Updated 1 month ago

About

Community (non-official) plugin wiring GBrain (garrytan/gbrain, MIT) into Hermes as a local memory layer over MCP: /gbrain status command, desktop starter, and per-profile Postgres setup/update/maintenance scripts. Not affiliated with upstream. Disclosure — the plugin itself only adds a /gbrain status command and a Desktop pane; GBrain is installed by the user via scripts/setup.sh, which installs bun (curl|bash), an unpinned gbrain from GitHub HEAD, and provisions PostgreSQL with sudo.

hermes-gbrain

CI
Security scans
License: MIT

Community (non-official) Hermes plugin that wires GBrain into Hermes as a memory
layer. It provides a unified package modeled on the pattern of one-link install,
per-profile setup, easy updates, and a small desktop starter UI.

GBrain itself is developed and maintained by garrytan
(garrytan/gbrain, MIT). This plugin is an independent community add-on: it is not
affiliated with, endorsed by, or maintained by the upstream GBrain project.

Upstream: garrytan/gbrain — see the Credits section below.

GBrain connects to Hermes over MCP (gbrain serve). This plugin adds:

  • a /gbrain slash command (quick status) on the CLI and gateway
  • a desktop UI starter (desktop/plugin.js)
  • scripts/setup.sh (fresh default / migrate, per-profile, source options)
  • scripts/update.sh (always pull the latest gbrain via self-upgrade)
  • docs/AGENT-GUIDE.md — the agent-facing guide (primary source, cloned with the repo)

Important: installing the plugin alone is not enough — the mcp_servers.gbrain
entry in the profile config is also required. setup.sh handles both.

Install (fresh VPS / migration)

For agents: follow docs/AGENT-GUIDE.md (read it once the plugin is cloned).
Manual (default profile, shared source):

# 1. install the plugin from git (--force: security scanner flags the bun-installer
#    subprocess scripts as trusted)
hermes plugins install kerrz2020/hermes-gbrain --enable --force
# 2. setup — runs the full path: bun → gbrain → Postgres+pgvector+grant → init keyless →
#    brain dir → source → writer claim → ACTIVATE → MCP wrapper → 3 crons → plugin → verify
bash ~/.hermes/plugins/gbrain-plugin/scripts/setup.sh --dry-run   # see the plan first
bash ~/.hermes/plugins/gbrain-plugin/scripts/setup.sh
# 3. restart the gateway from the user's shell

Default mode = KEYLESS on POSTGRES (gbrain init --url "postgresql://gbrain:***@127.0.0.1:5432/gbrain" --no-embedding --non-interactive):
no API key, no LLM, no local embeddings — just keyword FTS (tsvector) + memory.
Suited to fully-offline use. setup.sh provisions Postgres itself
(role/db/extension/grant/BYPASSRLS/v35 objects). Semantic upgrade later without
rework: gbrain init --force --embedding-model voyage:voyage-4.

Validation

bash scripts/verify-install-fresh.sh runs a drill in an isolated HOME sandbox with a
throwaway role/database: full provisioning path, 20 positive-state checks (including the
keyless posture and the keyless timeline path), then cleanup. Your real brain is never
touched. Expected values for every check are in docs/VALIDATION.md.

Per-profile & source

bash scripts/setup.sh --profile myprofile                 # a specific profile
bash scripts/setup.sh --source dedicated:work --profile friday  # isolated source (work brain)
bash scripts/setup.sh --migrate ~/brain                   # OPTIONAL: reuse an existing brain repo
bash scripts/setup.sh --engine pglite --profile myprofile # single-writer option (default: postgres)
  • shared (default): one federated source; writers split per agents/<name>/.
  • dedicated:: isolated source (federated=false), env GBRAIN_SOURCE=<name> on
    the MCP profile → reads/writes stay inside that source. Good for a separate domain brain.
  • Default migration is fresh-empty; --migrate <dir> syncs an existing brain repo.

Update (always latest)

bash scripts/update.sh    # gbrain self-upgrade + sync + hermes plugins update gbrain-plugin

Silent when already up to date → safe for unattended cron.

Structure

hermes-gbrain/
├── plugin.yaml          # manifest (v1-compat) + /gbrain command
├── __init__.py          # /gbrain status (fail-open)
├── desktop/plugin.js    # desktop UI starter (pane + chip)
├── docs/AGENT-GUIDE.md  # agent guide (primary source)
├── docs/VALIDATION.md   # what the fresh-install drill checks
├── tools/fm-check.py    # brain-frontmatter validator
├── scripts/setup.sh     # fresh/migrate/per-profile/source + PG provision + claim/activate + crons
├── scripts/update.sh    # self-upgrade + sync (--no-pull) + plugin update
└── scripts/maintenance/ # brain-sync.sh · gbrain-maintain.sh · gbrain-weekly-backup.sh (cron set)

Credits

  • GBrain — developed and maintained by garrytan;
    garrytan/gbrain, MIT licensed. Install it from GitHub, never from npm (npm has an
    unrelated package that shadows it): bun install -g github:garrytan/gbrain.
  • This plugin — a non-official community helper, MIT licensed.

README from the reviewed commit 51f5b55; it updates when the author re-pins.

More by kerrz2020

2 plugins by kerrz2020 in the catalog.