HermesFlow — Hermes Agent Docs

Community-driven, AI-elaborated documentation for Hermes Agent (Nous Research).

This is a living documentation project. Humans contribute raw work sessions. An AI agent (Kos) reads them daily, extracts knowledge, and produces structured, queryable documentation.

Sister project of openclaw-docs (which covers Claude Code / OpenClaw).

How It Works

Contributors write session logs → sessions/
                ↓
    Kos reads & analyzes daily (cron)
                ↓
    Structured docs generated → docs/
                ↓
    Index updated → docs/index.yaml
                ↓
    Anyone can fetch & query the docs

Quick Start

Read the docs

Browse docs/ by category:

Category	Path	Description
Getting Started	docs/getting-started/	Installation, first run, basic setup
Guides	docs/guides/	Step-by-step how-to for specific tasks (profiles, gateway, cron, integrations)
Reference	docs/reference/	Complete reference for CLI, profile commands, MCP, skills, tools
Concepts	docs/concepts/	Architecture, mental models, design decisions (multi-profile fleet, FTS5 sessions, Honcho user modeling)
Troubleshooting	docs/troubleshooting/	Known issues, errors, and solutions

Programmatic access

Fetch the index for machine-readable doc discovery:

# Full index
curl -s https://raw.githubusercontent.com/kos-domus/hermes-docs/main/docs/index.yaml

# Specific doc
curl -s https://raw.githubusercontent.com/kos-domus/hermes-docs/main/docs/guides/profile-multi-fleet.md

Use in your agent context

Add to your agent's instructions / SOUL.md:

# Hermes Reference
Fetch https://raw.githubusercontent.com/kos-domus/hermes-docs/main/docs/index.yaml
for the documentation index. Fetch individual docs by path as needed.

Contributing

We welcome session contributions from anyone using Hermes Agent. See CONTRIBUTING.md for details.

TL;DR:

1. Fork the repo
2. Copy sessions/_template.md to sessions/YYYY-MM-DD-your-topic.md
3. Fill in your session details
4. Set status: ready
5. Open a PR

Kos will process your session in the next daily run and integrate your knowledge into the docs.

Project Structure

hermes-docs/
├── CLAUDE.md              # Instructions for Kos (elaboration engine)
├── sessions/              # Raw work session logs (input)
│   ├── _template.md       # Template for new sessions
│   └── schema.yaml        # Validation schema
├── docs/                  # Elaborated documentation (output)
│   ├── index.yaml         # Machine-readable index (fetch this first)
│   ├── getting-started/
│   ├── guides/
│   ├── reference/
│   ├── concepts/
│   └── troubleshooting/
├── strategy/              # Hermes strategy / architecture decisions
├── skill-draft/           # Skill drafts before promotion to Hermes catalog
└── changelog/
    └── CHANGELOG.md       # Log of all elaboration runs

Why a separate repo from openclaw-docs?

OpenClaw (Claude Code) and Hermes (Nous Research) are different runtimes with different idioms:

• OpenClaw: workspace-based, custom YAML config, OpenClaw-specific tool registry
• Hermes: profile-based, FTS5 session search, Honcho user modeling, multi-platform messaging gateway built-in, agentskills.io standard

Keeping the docs separate avoids confusion and lets each repo follow its native conventions.

License

MIT — use freely, contribute back.