Self-host

The community edition is the full core, free: the author, flows, MCP, and hybrid search. Run it on your own hardware and keep your keys.

What’s in the community edition

The document author and knowledge graph
Flows and capture
The MCP endpoint (every AI client connects to your instance)
Hybrid (keyword + semantic) search

The hosted graph and meeting service is the licensed part — see Editions.

Quickstart (~4 minutes)

The core ships as a Docker Compose stack.

git clone https://github.com/nbkdoesntknowcoding/mnema
cd mnema/infra
cp .env.selfhost.example .env   # set your keys
docker compose -f docker-compose.selfhost.yml up -d

Your instance comes up with the app, the API, and the MCP endpoint. Point your AI clients at your own origin instead of the hosted one:

http://localhost:8080/mcp

Put it behind a reverse proxy (nginx/Caddy) for TLS and a public hostname; forward the Upgrade and Connection headers so the MCP stream works.

Dev mode on self-host

The entire dev pillar is part of the free core and runs on a self-hosted community instance with no licence:

Sessions, Cost and Optimize — agent capture, spend and findings
Projects & board — Kanban, projects and sprints
the coding-agent hook and the full multi-AI connectivity surface (REST API, OpenAPI, personal API keys)

Point the hook installer at your own origin instead of the hosted one:

MNEMA_API_ORIGIN=https://your-host MNEMA_HOOK_TOKEN=<token> MNEMA_WORKSPACE_ID=<id> \
  bash <(curl -sf https://your-host/install/claude-hooks.sh)

What needs a licence is only the understanding layer — the knowledge graph over your sessions and decisions, cross-repo rollups, and meetings — see Editions.

For the licence and what you can do with the core, see Licence.