100yenadmin

The official Fish Audio plugin, maintained by Electric Sheep: expressive text-to-speech in 80+ languages, speech-to-text, Fish Audio's community voice library, voice cloning and voice design, with a Voices page in Hermes Desktop. Fish Audio becomes a Hermes speech and transcription provider for voice replies, read-aloud, voice notes and Desktop dictation, with sentence-by-sentence streaming in CLI voice mode and Desktop, and live dictation captions on Hermes 0.21.6+ (stt.streaming: true). Adds the fish_speak, fish_voices and fish_transcribe tools, /fish and hermes fish commands, and three skills. Disclosure — sends your Fish API key and the text, audio (with file names) and voice titles, tags, descriptions, cover images and design instructions you ask it to process to api.fish.audio, or to the base_url you configure, and reads your Fish wallet for balances and model choice; with live dictation on, microphone audio streams to api.fish.audio's realtime speech-to-text endpoint while you speak, billed per committed phrase rounded up to a second; synthesis, transcription, cloning and voice design are paid requests billed to your Fish account, except the free s2.1-pro-free model, whose requests Fish may use to improve its models. The key comes from the profile's secret scope (Hermes stores it in the profile .env; a single-profile install may also read FISH_API_KEY from the environment). Generated audio and subtitles are written to Hermes's audio cache; clone samples and recordings it uploads must be regular audio files, and Hermes config, secret and credential paths are refused. Cloning and deleting voices from chat or the model tools go through Hermes's approval gate and follow your approval settings. The Desktop Voices page and a Fish Audio settings page add gateway routes under /api/plugins/fish-audio/, behind the Hermes dashboard's existing authentication. On the Voices page, previews, clones and voice designs are billed; cloning a voice or saving a designed one creates a voice in your Fish account; cloning and deleting are confirmed on the page (a speaker-consent box, a typed voice name) instead of through Hermes's approval gate; and Use sets the profile's Fish voice, and its speech provider when none is set. The settings page and /fish alias can set or reset the profile's Fish model and voice and save up to 50 voice names in the profile's config. Preview and design audio files are deleted from the gateway once returned; a design's audio is held in gateway memory so you can save it, until it is saved, a later design or save finds it over an hour old, or the gateway stops; clone uploads are deleted after the clone, or by the next upload once idle for 15 minutes. With a key set, Desktop reads your Fish wallet when an agent is selected and about every five minutes, for the status-bar credit; favourites stay in Desktop's plugin storage on that computer, and fetched previews stay in the window's memory until it closes. No telemetry.

Named subagent teammates for Hermes: the operator defines a roster (instructions, toolsets, model), and the agent assigns them work, checks results, steers a running teammate, follows up with a digest of earlier results, and can work a Kanban teammates lane with a review handoff. Launches subagents only through the public ctx.subagent_lifecycle API. Stock Hermes; no patched core. Disclosure — adds five tools and a /teammates command; the model picks only a teammate name, and the operator's config decides what it means. Teammates run as in-process Hermes subagents on the parent's provider credentials; toolsets narrow the parent's tools at toolset granularity only (the file toolset includes write_file and patch), and a teammate with invalid settings refuses to launch. A delegated child cannot start teammates. Steering dispatches delegate_task's steer action with the turn-bound parent, so under plugins.isolation: host every tool except the roster refuses. It keeps a per-profile SQLite run ledger in its plugin-data folder (goals, statuses, clipped summaries; no transcripts or credentials). For a card assigned to its non-profile lane, it claims the card through the Kanban DB, heartbeats the claim and its worker, then moves the card to review without a reviewer, or blocks it on failure, a stop or an unavailable result; it never completes a card. It runs one in-process monitor thread per lane run to keep the claim alive; no network calls, child processes or self-updates of its own.

Claude Code-style plan mode on every interactive Hermes surface: /plan or /planmode on blocks mutating Hermes tools dispatched through pre_tool_call (writes outside .hermes/plans, terminal, code execution, delegation, MCP) while the agent explores, asks clarifying questions and writes a plan. The agent submits the plan with plan_mode(submit) and you approve it through Hermes' own approval prompt (gateway buttons, the TUI/Desktop card, the CLI panel or /approve); it then implements in the same turn and tracks todo_list progress. Deny keeps planning and the agent revises. When yolo or approvals.mode: off would auto-approve, it asks one clarify question instead (Approve plan rev N / Keep planning), and only that answer approves. By default plans are compact and decision-complete and the agent does not commit unless asked (plan_style, allow_commits and plan_skill settings; plan_style: core keeps the zero-context /plan craft). The default lean profile keeps compact planning short: a brief planning note, at most three targeted questions, and no todo bookkeeping for short plans (plan_profile: classic keeps 0.3.5); on TUI, Desktop and hermes serve the agent writes the plan once, in its reply, and the plugin saves it. Classic CLI enforced on every supported Hermes; gateway, TUI and Desktop from Hermes v2026.9.24 (0.21.5), and refused on older builds. Disclosure — plan-mode does not enforce Codex app-server exec/applyPatch or TUI /background and btw side agents. It reads eight groups of internal Hermes seams for session identity, workspace, profile and skills.inline_shell, each guarded with a documented fallback. It creates .hermes/plans under the session working directory and keeps per-session state in plugin state. It adds a turn note while planning or executing, a ≤200-char system-prompt hint, and a one-line footer to short replies on chat platforms. When you pick "Always" on a plan approval, Hermes core writes a plugin_rule:plan-mode entry to command_allowlist in config.yaml. With allow_gateway_injection: true it queues one message to start work after a typed /planmode approve. On TUI, Desktop and hermes serve sessions (lean profile, plan_once) it reads the visible reply text from post_api_request while planning and saves its <proposed_plan> block as the plan file. No network calls, no subprocesses, no self-updater.

A live board the agent builds for you: a Board tab in the Hermes dashboard and a Desktop page where the agent lays out tabs and widgets (notes, markdown, tables, charts, KPI cards, live Hermes data, approvals) through 19 native boardstate_* tools. Runs a local Node >= 20 sidecar on loopback with a per-spawn nonce; no network access unless you opt in to operator-gated connectors. Disclosure — spawns a local Node.js ≥20 sidecar (from PATH/HERMES_NODE_BIN, nothing downloaded) on 127.0.0.1 on first use; state under HERMES_HOME/boardstate-state/. Reads the dashboard's loopback session token from Hermes internals (web_server._SESSION_TOKEN) and passes it to the sidecar for live usage/sessions/cron widgets; mounts one token-gated static route for approved widget assets. No outbound network unless you author boardstate.connectors.json (stdio/HTTP MCP connectors); connector actions need operator approval.

Wakes and watchers for Hermes agents, maintained by Electric Sheep. You (with /pulse) or your agent (with the pulse tool) arm follow-ups at a time, recurring checks, a check-in after a long silence, or deterministic watchers on a file, process, URL, read-only command or Hermes event (kanban, subagent, API error, reactions). A wake is a turn in the same messaging DM that armed it, on that chat's own model: no side model, cron session or subagent. Watchers decide without a model call; the agent answers each wake with a structured verdict (notify or stay silent, never a typed sentinel); a no-op governor slows, then pauses pointless wakes and asks once. Every guardrail is a user setting (quiet hours, daily caps, unanswered mute, floors, lifetimes) that the agent works inside and can never widen. v0.1 delivers to messaging-gateway DMs only and needs plugins.entries.heartbeat2.allow_gateway_injection: true. Disclosure — wakes and watchers run full turns on the conversation's own model, billed like any message, at your conversation's context size, with the conversation's normal tools and approval prompts; the agent can arm wakes and watchers itself unless you set agent_can_create to ask_first or no; command watchers run allowlisted read-only commands (default gh pr view/checks, gh run view/list, git ls-remote/status/log) without an approval prompt, with no shell, a sanitized environment (no provider keys, GH_TOKEN or GITHUB_TOKEN) and the process group killed after 30 s, can use your stored git/gh credentials, and `git ls-remote <url>` / `gh -R` can contact any host your machine can reach; URL watchers make the GET requests you or the agent configured, without URL credentials, cookies, auth headers or proxies, always refuse cloud-metadata addresses and refuse private or local addresses unless url_allow_private is on; file watchers refuse paths Hermes blocks from reading; notes, summaries and event payloads are stored in plaintext in the profile's plugin-data; no telemetry.

Cloud Files for Hermes Desktop — browse, search, create folders and bulk-upload files and whole folders to the machine your agent runs on, read and edit its Markdown and text files (view first; nothing is written until you review the line diff and press Save), import files from the agent's own Google Drive, and hand the agent file locations from the chat + menu without re-uploading. Built for remote gateways (token or sign-in); works locally too. Disclosure — adds a file API to the gateway (dashboard plugin routes under the gateway's existing auth) over the agent's working folders, by default terminal.cwd or else the gateway user's home with its dot-entries hidden. It lists and searches those folders, reads Markdown and text files, creates folders and writes uploads and Drive imports (staged as hidden .cfm-* temp files in the destination folder; a name clash keeps both files). It overwrites only an existing .md/.markdown/.txt file you open and explicitly Save in its editor (refused if the file changed since you opened it). It never deletes your files (it removes only its own stale or aborted .cfm-* temp files and staging folders), never follows links out of those folders, and never lists, reads or writes the agent's Hermes home apart from a working folder set inside it (roots, terminal.cwd, or a workspace/ folder it creates when the working folder would be the home itself). Its only network use is Google Drive, read-only, through the agent's own google-workspace skill, whose scripts (from skills/productivity/google-workspace in the agent's Hermes home, else the copy bundled with Hermes) it runs with the gateway's Python and a minimal environment without API keys; the plugin never reads the Google token. While Desktop is open it re-checks Drive sign-in about once a minute; otherwise it makes no network calls.