🇰🇷 한국어
A personal knowledge pipeline that turns your Slack conversations and Claude Code sessions into a self-organizing, searchable Obsidian vault — automatically summarized, linked, and cross-referenced by an LLM.
Point it at a Slack workspace and/or your Claude Code sessions, and it keeps a living wiki of projects, people, and concepts up to date on its own — no manual note-taking.
- Multi-source ingest — Slack channels and Claude Code session transcripts today; new sources plug in with one small provider
- Automatic knowledge graph — entities (projects, people, concepts), keywords, and cross-links extracted and maintained by an LLM
- Self-updating wiki — Markdown pages generated and kept current as new conversations come in
- Incremental processing — only reprocesses what changed
- LLM response cache — avoids redundant API calls
- Watch & daemon modes — runs continuously on a schedule, unattended
- Knowledge evaluation — health/learning scores, coverage gaps, retrieval benchmarking
- MCP server — query your vault directly from Claude (search, context building, health checks)
# 1. Clone and set up a virtual environment
git clone https://github.com/ginaseo/hermes-knowledge-engine.git
cd hermes-knowledge-engine
python -m venv .venv
.venv\Scripts\activate # Windows
# source .venv/bin/activate # Linux/macOS
# 2. Install (exposes the `hermes` CLI command)
pip install .
# 3. Configure environment variables
cp .env.example .env
# Set HERMES_API_URL / HERMES_API_KEY (LLM endpoint) and, if using Slack, SLACK_BOT_TOKEN / SLACK_CHANNEL_IDS
# Or set HERMES_LOCAL_HEURISTIC=1 to skip the LLM entirely (free, lower-quality summaries/entities)
# 4. Run the pipeline
hermes runSee INSTALL.md for full setup, environment variables, and dev setup.
hermes # full pipeline
hermes watch # poll for changes every 30s
hermes daemon # continuously scheduled job runner
hermes evaluate # knowledge stats, quality, health & learning scores
hermes validate # validate vault integrity
hermes --help # everything else- INSTALL.md — full installation, configuration, and dev setup
- docs/pipeline.md — processor pipeline, folder structure, caching/incremental details
- docs/ingest-sources.md — Slack & Claude Code providers, adding a new source
- docs/operations.md — daemon mode, evaluation, retrieval benchmark, deployment notes
- docs/vault-sync.md — mirroring the vault across machines with Syncthing
- MCP.md — MCP server setup and tools
- ARCHITECTURE.md — system design