mirror of https://github.com/0xrsydn/nix-hermes-agent.git synced 2026-08-07 09:03:52 +00:00

Ciphercat 010a1b6644 feat: bump hermes-agent 0.2.0 → 0.3.0 + auto-update CI

- Update package.nix to v0.3.0 (NousResearch/hermes-agent v2026.3.17)
- Add scripts/update-pins.sh — checks latest stable release, prefetches, validates build
- Add .github/workflows/update-pins.yml — runs every 6h + manual dispatch
- Pattern follows nix-openclaw's yolo-update approach

2026-03-17 19:10:01 +00:00

12 KiB

Raw Blame History

Hermes Agent Configuration Map

Source: hermes_cli/config.py + gateway/config.py (v0.2.0)

Everything below can be set declaratively via the NixOS module's services.hermes-agent.config option (rendered as cli-config.yaml), environmentFiles (secrets), or environment (non-secret env vars).

1. Model & Provider

Key	Type	Default	Description
`model`	string	`"anthropic/claude-opus-4.6"`	Default model (`provider/model` format)
`toolsets`	list	`["hermes-cli"]`	Enabled toolsets (e.g. `["all"]`, `["hermes-cli", "browser"]`)

Auth providers (resolved automatically based on env vars / OAuth):

Nous Portal (OAuth) → hermes login --provider nous
OpenAI Codex (OAuth) → hermes login --provider openai-codex
OpenRouter → OPENROUTER_API_KEY
Z.AI / GLM → GLM_API_KEY / ZAI_API_KEY
Kimi → KIMI_API_KEY
MiniMax → MINIMAX_API_KEY
Anthropic → ANTHROPIC_API_KEY or Claude Code OAuth

2. Agent Behavior

Key	Type	Default	Description
`agent.max_turns`	int	`90`	Max tool-calling iterations per conversation

3. Terminal / Sandbox

Key	Type	Default	Description
`terminal.backend`	string	`"local"`	`local` / `docker` / `singularity` / `modal` / `daytona` / `ssh`
`terminal.cwd`	string	`"."`	Working directory for commands
`terminal.timeout`	int	`180`	Command timeout (seconds)
`terminal.docker_image`	string	`"nikolaik/python-nodejs:..."`	Docker image
`terminal.singularity_image`	string	...	Singularity image
`terminal.modal_image`	string	...	Modal image
`terminal.daytona_image`	string	...	Daytona image
`terminal.container_cpu`	int	`1`	Container CPU limit
`terminal.container_memory`	int	`5120`	Container memory (MB)
`terminal.container_disk`	int	`51200`	Container disk (MB)
`terminal.container_persistent`	bool	`true`	Persist filesystem across sessions
`terminal.docker_volumes`	list	`[]`	Docker volume mounts (`host:container`)

SSH backend env vars:

TERMINAL_SSH_HOST, TERMINAL_SSH_USER, TERMINAL_SSH_KEY

4. Context Compression

Key	Type	Default	Description
`compression.enabled`	bool	`true`	Auto-compress long conversations
`compression.threshold`	float	`0.50`	Context usage % to trigger compression
`compression.summary_model`	string	`"google/gemini-3-flash-preview"`	Model for summaries
`compression.summary_provider`	string	`"auto"`	Provider for compression model

5. Auxiliary Models

Side-task models (vision, web extraction, etc). Each has the same schema:

Key	Type	Default	Description
`auxiliary.<task>.provider`	string	`"auto"`	Provider override
`auxiliary.<task>.model`	string	`""`	Model override
`auxiliary.<task>.base_url`	string	`""`	Direct endpoint URL
`auxiliary.<task>.api_key`	string	`""`	API key for endpoint

Tasks: vision, web_extract, compression, session_search, skills_hub, mcp, flush_memories

6. Delegation (Subagents)

Key	Type	Default	Description
`delegation.model`	string	`""`	Model for subagents (empty = inherit parent)
`delegation.provider`	string	`""`	Provider for subagents
`delegation.base_url`	string	`""`	Direct endpoint for subagents
`delegation.api_key`	string	`""`	API key for delegation endpoint

7. Display

Key	Type	Default	Description
`display.compact`	bool	`false`	Compact output mode
`display.personality`	string	`"kawaii"`	UI personality
`display.resume_display`	string	`"full"`	Session resume display
`display.bell_on_complete`	bool	`false`	Bell sound on completion
`display.show_reasoning`	bool	`false`	Show model reasoning
`display.skin`	string	`"default"`	UI skin

8. Memory

Key	Type	Default	Description
`memory.memory_enabled`	bool	`true`	Persistent curated memory
`memory.user_profile_enabled`	bool	`true`	User profile memory
`memory.memory_char_limit`	int	`2200`	Max chars for memory context
`memory.user_char_limit`	int	`1375`	Max chars for user profile

9. Browser

Key	Type	Default	Description
`browser.inactivity_timeout`	int	`120`	Browser auto-close (seconds)
`browser.record_sessions`	bool	`false`	Record browser sessions as WebM

10. Checkpoints (Filesystem Snapshots)

Key	Type	Default	Description
`checkpoints.enabled`	bool	`false`	Auto-snapshot before destructive ops
`checkpoints.max_snapshots`	int	`50`	Max checkpoints per directory

11. TTS (Text-to-Speech)

Key	Type	Default	Description
`tts.provider`	string	`"edge"`	`edge` (free) / `elevenlabs` (premium) / `openai`
`tts.edge.voice`	string	`"en-US-AriaNeural"`	Edge TTS voice
`tts.elevenlabs.voice_id`	string	`"pNInz6obpgDQGcFmaJgB"`	ElevenLabs voice ID
`tts.elevenlabs.model_id`	string	`"eleven_multilingual_v2"`	ElevenLabs model
`tts.openai.model`	string	`"gpt-4o-mini-tts"`	OpenAI TTS model
`tts.openai.voice`	string	`"alloy"`	OpenAI TTS voice

12. STT (Speech-to-Text)

Key	Type	Default	Description
`stt.enabled`	bool	`true`	Enable voice transcription
`stt.provider`	string	`"local"`	`local` (faster-whisper) / `groq` / `openai`
`stt.local.model`	string	`"base"`	Local model size: tiny/base/small/medium/large-v3
`stt.openai.model`	string	`"whisper-1"`	OpenAI STT model

13. Voice (Interactive)

Key	Type	Default	Description
`voice.record_key`	string	`"ctrl+b"`	Push-to-talk key
`voice.max_recording_seconds`	int	`120`	Max recording length
`voice.auto_tts`	bool	`false`	Auto-speak responses
`voice.silence_threshold`	int	`200`	RMS silence threshold
`voice.silence_duration`	float	`3.0`	Seconds of silence → auto-stop

14. Human Delay (Anti-Detection)

Key	Type	Default	Description
`human_delay.mode`	string	`"off"`	`off` / `on` / `adaptive`
`human_delay.min_ms`	int	`800`	Min delay (ms)
`human_delay.max_ms`	int	`2500`	Max delay (ms)

15. Security

Key	Type	Default	Description
`security.redact_secrets`	bool	`true`	Redact API keys from tool output
`security.tirith_enabled`	bool	`true`	Pre-exec scanning via tirith
`security.tirith_path`	string	`"tirith"`	Path to tirith binary
`security.tirith_timeout`	int	`5`	Tirith scan timeout (seconds)
`security.tirith_fail_open`	bool	`true`	Allow exec if tirith fails

16. Discord

Key	Type	Default	Description
`discord.require_mention`	bool	`true`	Require @mention to respond in channels
`discord.free_response_channels`	string	`""`	Comma-separated channel IDs for free response
`discord.auto_thread`	bool	`true`	Auto-create threads on @mention

17. Session Reset Policy

Key	Type	Default	Description
`session_reset.mode`	string	`"both"`	`daily` / `idle` / `both` / `none`
`session_reset.at_hour`	int	`4`	Daily reset hour (0-23, local time)
`session_reset.idle_minutes`	int	`1440`	Idle timeout before reset (minutes)

18. Miscellaneous

Key	Type	Default	Description
`timezone`	string	`""`	IANA timezone (empty = server-local)
`command_allowlist`	list	`[]`	Permanently allowed dangerous commands
`quick_commands`	dict	`{}`	User-defined slash commands (exec type)
`personalities`	dict	`{}`	Custom personality prompts
`prefill_messages_file`	string	`""`	Path to JSON prefill messages
`honcho`	dict	`{}`	Honcho AI-native memory overrides

19. MCP Servers

Defined via services.hermes-agent.mcpServers in the NixOS module:

mcpServers = {
  my-server = {
    command = "npx";
    args = [ "-y" "my-mcp-server" ];
    env = { API_KEY = "..."; };
    timeout = 30;
  };
};

Merged into config.mcp_servers in the rendered YAML.

20. Fallback Model (Failover)

Not in defaults — must be explicitly configured:

fallback_model:
  provider: openrouter     # openrouter | nous | openai-codex | zai | kimi-coding | minimax
  model: anthropic/claude-sonnet-4
  # Optional for custom endpoints:
  # base_url: "http://..."
  # api_key_env: "MY_CUSTOM_KEY"

Triggers on: 429 (rate limit), 529 (overload), 503 (service error), connection failures.

21. Environment Variables (Secrets via `environmentFiles`)

Provider Keys

Variable	Provider
`OPENROUTER_API_KEY`	OpenRouter (default fallback)
`GLM_API_KEY` / `ZAI_API_KEY`	Z.AI / GLM
`KIMI_API_KEY`	Kimi / Moonshot
`MINIMAX_API_KEY`	MiniMax (international)
`MINIMAX_CN_API_KEY`	MiniMax (China)
`ANTHROPIC_API_KEY`	Anthropic direct
`ANTHROPIC_TOKEN`	Anthropic OAuth/setup token

Tool Keys

Variable	Tool
`FIRECRAWL_API_KEY`	Web search/scraping
`BROWSERBASE_API_KEY`	Cloud browser
`FAL_KEY`	Image generation
`ELEVENLABS_API_KEY`	Premium TTS
`VOICE_TOOLS_OPENAI_KEY`	Whisper STT + OpenAI TTS
`GITHUB_TOKEN`	Skills Hub
`HONCHO_API_KEY`	AI-native memory
`TINKER_API_KEY`	RL training
`WANDB_API_KEY`	Experiment tracking

Messaging Tokens

Variable	Platform
`TELEGRAM_BOT_TOKEN`	Telegram
`TELEGRAM_ALLOWED_USERS`	Telegram user whitelist
`DISCORD_BOT_TOKEN`	Discord
`DISCORD_ALLOWED_USERS`	Discord user whitelist
`SLACK_BOT_TOKEN`	Slack
`SLACK_APP_TOKEN`	Slack (Socket Mode)
`SIGNAL_HTTP_URL`	Signal
`SIGNAL_ACCOUNT`	Signal
`HASS_TOKEN`	Home Assistant
`EMAIL_ADDRESS` / `EMAIL_PASSWORD` / `EMAIL_IMAP_HOST` / `EMAIL_SMTP_HOST`	Email
`GATEWAY_ALLOW_ALL_USERS`	Allow all users (global)

Agent Settings

Variable	Description
`MESSAGING_CWD`	Working directory for messaging
`SUDO_PASSWORD`	Sudo password for terminal
`HERMES_MAX_ITERATIONS`	Max iterations override

22. Custom Providers (config.yaml)

custom_providers:
  - name: "local-vllm"
    base_url: "http://localhost:8000/v1"
    api_key: ""  # optional

Use with: hermes chat --provider custom:local-vllm

Summary: What Goes Where

What	Where	NixOS Module Option
Model, toolsets, display, terminal, etc.	`cli-config.yaml`	`config = { ... }`
API keys, tokens, passwords	`.env` / env vars	`environmentFiles = [ "/run/secrets/hermes.env" ]`
OAuth tokens (Nous, Codex)	`auth.json`	Runtime via `hermes login` (persistent)
MCP servers	`cli-config.yaml`	`mcpServers = { ... }`
SOUL.md, AGENTS.md, USER.md	Workspace files	`documents = { ... }`
Gateway platform config	env vars or `gateway.json`	`environmentFiles`

12 KiB Raw Blame History