mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-05-06 10:47:12 +08:00
53f875d063
Skips the ~145 ms AST scan of every tools/*.py file at every startup.
CLI cold import drops ~100 ms on a machine where Phase 1's other
lazy-imports already landed.
## What changed
tools/registry.py:_load_manifest() reads a committed tuple of tool
module names from tools/_manifest.py. discover_builtin_tools() tries
the manifest first; falls back to the AST scan only when the manifest
is missing (fresh checkout, or the generator hasn't been run yet) or
when the caller passes a non-default tools_dir (tests, embedders).
scripts/build_tool_manifest.py regenerates the manifest. Run manually
after adding a new tools/*.py, or let CI catch drift.
.github/workflows/tests.yml runs the generator with --check on every
PR. Rejects commits that add/remove a self-registering tool module
without updating the manifest.
tools/_manifest.py is the committed artifact, 27 built-in modules.
## Design decisions
No runtime mtime-drift check. Considered adding a stat walk of
tools/*.py at startup to auto-invalidate the manifest on dev machines,
rejected because:
- Adds overhead to the path we're trying to speed up (66 stats,
~5-10 ms worst case, defeats part of the savings).
- Gives false positives when devs edit tools/ helpers (ansi_strip,
approval, budget_config, path_security, etc.) that don't call
registry.register() — those edits wouldn't require a regen.
- CI check is strictly better: catches the real error class (manifest
missing a self-registering tool) without runtime cost, and developers
get feedback at PR time instead of silent divergence.
Manifest coverage: built-in tools/*.py ONLY. Plugin tools (via
ctx.register_tool) and MCP tools (registered dynamically on server
connect) use separate registration paths that this change doesn't
touch. End-user-visible behavior: unchanged — installing a plugin or
connecting an MCP server works exactly as before.
## Failure modes
1. Manifest lists a tool that no longer exists → importlib.import_module
raises, logged as warning, skipped. Non-fatal but noisy. CI check
catches this class before merge.
2. Manifest missing a new tool → the tool isn't registered at startup.
User-visible. CI --check catches before merge.
3. Manifest missing entirely (tools/_manifest.py deleted) → fallback to
AST scan, startup gets slower by ~145 ms but behavior is identical.
## Measurements
discover_builtin_tools() isolated:
without manifest (AST scan): 674 ms min, 702 ms median
with manifest: 524 ms min, 550 ms median
savings: 150 ms
Full 'import cli':
before (main): ~950 ms
after (this PR): ~846 ms
savings: ~100 ms
(The isolated savings is bigger than the cli delta because some AST
scan overhead was being double-paid via transitive imports; manifest
removes it once at the source.)
## Validation
- 121/121 tests/tools/test_registry + test_terminal_tool +
test_skills_tool pass (unchanged from main).
- 68 tools registered with manifest (matches baseline).
- CI --check flag verified: passes for the committed manifest, fails
with exit 1 when I drop a new tool into tools/ without regenerating.
- Live hermes chat smoke: 1 turn + /quit, agent successfully enumerated
all registered tools when asked to list them. 36-line log window,
zero errors.
Hermes Agent ☤
The self-improving AI agent built by Nous Research. It's the only agent with a built-in learning loop — it creates skills from experience, improves them during use, nudges itself to persist knowledge, searches its own past conversations, and builds a deepening model of who you are across sessions. Run it on a $5 VPS, a GPU cluster, or serverless infrastructure that costs nearly nothing when idle. It's not tied to your laptop — talk to it from Telegram while it works on a cloud VM.
Use any model you want — Nous Portal, OpenRouter (200+ models), NVIDIA NIM (Nemotron), Xiaomi MiMo, z.ai/GLM, Kimi/Moonshot, MiniMax, Hugging Face, OpenAI, or your own endpoint. Switch with hermes model — no code changes, no lock-in.
| A real terminal interface | Full TUI with multiline editing, slash-command autocomplete, conversation history, interrupt-and-redirect, and streaming tool output. |
| Lives where you do | Telegram, Discord, Slack, WhatsApp, Signal, and CLI — all from a single gateway process. Voice memo transcription, cross-platform conversation continuity. |
| A closed learning loop | Agent-curated memory with periodic nudges. Autonomous skill creation after complex tasks. Skills self-improve during use. FTS5 session search with LLM summarization for cross-session recall. Honcho dialectic user modeling. Compatible with the agentskills.io open standard. |
| Scheduled automations | Built-in cron scheduler with delivery to any platform. Daily reports, nightly backups, weekly audits — all in natural language, running unattended. |
| Delegates and parallelizes | Spawn isolated subagents for parallel workstreams. Write Python scripts that call tools via RPC, collapsing multi-step pipelines into zero-context-cost turns. |
| Runs anywhere, not just your laptop | Six terminal backends — local, Docker, SSH, Daytona, Singularity, and Modal. Daytona and Modal offer serverless persistence — your agent's environment hibernates when idle and wakes on demand, costing nearly nothing between sessions. Run it on a $5 VPS or a GPU cluster. |
| Research-ready | Batch trajectory generation, Atropos RL environments, trajectory compression for training the next generation of tool-calling models. |
Quick Install
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
Works on Linux, macOS, WSL2, and Android via Termux. The installer handles the platform-specific setup for you.
Android / Termux: The tested manual path is documented in the Termux guide. On Termux, Hermes installs a curated
.[termux]extra because the full.[all]extra currently pulls Android-incompatible voice dependencies.Windows: Native Windows is not supported. Please install WSL2 and run the command above.
After installation:
source ~/.bashrc # reload shell (or: source ~/.zshrc)
hermes # start chatting!
Getting Started
hermes # Interactive CLI — start a conversation
hermes model # Choose your LLM provider and model
hermes tools # Configure which tools are enabled
hermes config set # Set individual config values
hermes gateway # Start the messaging gateway (Telegram, Discord, etc.)
hermes setup # Run the full setup wizard (configures everything at once)
hermes claw migrate # Migrate from OpenClaw (if coming from OpenClaw)
hermes update # Update to the latest version
hermes doctor # Diagnose any issues
CLI vs Messaging Quick Reference
Hermes has two entry points: start the terminal UI with hermes, or run the gateway and talk to it from Telegram, Discord, Slack, WhatsApp, Signal, or Email. Once you're in a conversation, many slash commands are shared across both interfaces.
| Action | CLI | Messaging platforms |
|---|---|---|
| Start chatting | hermes |
Run hermes gateway setup + hermes gateway start, then send the bot a message |
| Start fresh conversation | /new or /reset |
/new or /reset |
| Change model | /model [provider:model] |
/model [provider:model] |
| Set a personality | /personality [name] |
/personality [name] |
| Retry or undo the last turn | /retry, /undo |
/retry, /undo |
| Compress context / check usage | /compress, /usage, /insights [--days N] |
/compress, /usage, /insights [days] |
| Browse skills | /skills or /<skill-name> |
/<skill-name> |
| Interrupt current work | Ctrl+C or send a new message |
/stop or send a new message |
| Platform-specific status | /platforms |
/status, /sethome |
For the full command lists, see the CLI guide and the Messaging Gateway guide.
Documentation
All documentation lives at hermes-agent.nousresearch.com/docs:
| Section | What's Covered |
|---|---|
| Quickstart | Install → setup → first conversation in 2 minutes |
| CLI Usage | Commands, keybindings, personalities, sessions |
| Configuration | Config file, providers, models, all options |
| Messaging Gateway | Telegram, Discord, Slack, WhatsApp, Signal, Home Assistant |
| Security | Command approval, DM pairing, container isolation |
| Tools & Toolsets | 40+ tools, toolset system, terminal backends |
| Skills System | Procedural memory, Skills Hub, creating skills |
| Memory | Persistent memory, user profiles, best practices |
| MCP Integration | Connect any MCP server for extended capabilities |
| Cron Scheduling | Scheduled tasks with platform delivery |
| Context Files | Project context that shapes every conversation |
| Architecture | Project structure, agent loop, key classes |
| Contributing | Development setup, PR process, code style |
| CLI Reference | All commands and flags |
| Environment Variables | Complete env var reference |
Migrating from OpenClaw
If you're coming from OpenClaw, Hermes can automatically import your settings, memories, skills, and API keys.
During first-time setup: The setup wizard (hermes setup) automatically detects ~/.openclaw and offers to migrate before configuration begins.
Anytime after install:
hermes claw migrate # Interactive migration (full preset)
hermes claw migrate --dry-run # Preview what would be migrated
hermes claw migrate --preset user-data # Migrate without secrets
hermes claw migrate --overwrite # Overwrite existing conflicts
What gets imported:
- SOUL.md — persona file
- Memories — MEMORY.md and USER.md entries
- Skills — user-created skills →
~/.hermes/skills/openclaw-imports/ - Command allowlist — approval patterns
- Messaging settings — platform configs, allowed users, working directory
- API keys — allowlisted secrets (Telegram, OpenRouter, OpenAI, Anthropic, ElevenLabs)
- TTS assets — workspace audio files
- Workspace instructions — AGENTS.md (with
--workspace-target)
See hermes claw migrate --help for all options, or use the openclaw-migration skill for an interactive agent-guided migration with dry-run previews.
Contributing
We welcome contributions! See the Contributing Guide for development setup, code style, and PR process.
Quick start for contributors — clone and go with setup-hermes.sh:
git clone https://github.com/NousResearch/hermes-agent.git
cd hermes-agent
./setup-hermes.sh # installs uv, creates venv, installs .[all], symlinks ~/.local/bin/hermes
./hermes # auto-detects the venv, no need to `source` first
Manual path (equivalent to the above):
curl -LsSf https://astral.sh/uv/install.sh | sh
uv venv venv --python 3.11
source venv/bin/activate
uv pip install -e ".[all,dev]"
scripts/run_tests.sh
RL Training (optional): The RL/Atropos integration (
environments/) ships via theatroposlibandtinkerdependencies pulled in by.[all,dev]— no submodule setup required.
Community
- 💬 Discord
- 📚 Skills Hub
- 🐛 Issues
- 🔌 HermesClaw — Community WeChat bridge: Run Hermes Agent and OpenClaw on the same WeChat account.
License
MIT — see LICENSE.
Built by Nous Research.