dayyanj/nmem-studio repository overview

⁠nmem-studio

Pull an image, open a web page, and have a live cognitive agent in ~10 minutes — no config files, no PhD in the stack. nmem-studio is the onboarding + management UX over the nmem⁠ cognitive-memory engine: 6-tier memory, a symbol-graph world model, optional drives / goals / pursuit, grounded chat, and a live 3D "brain" you can watch think.

⁠Quick start — one line

docker run -p 127.0.0.1:8080:8080 -p 127.0.0.1:5174:5174 -v nmem:/data docker.io/dayyanj/nmem-studio

Open http://localhost:8080⁠ for the setup wizard, and http://localhost:5174⁠ for the brain viz. That single container is the whole appliance — Postgres + pgvector, the studio, and the viz run inside it, with all state on the nmem volume, so your agent survives docker rm and comes back.

⁠What you do in the wizard

1. Pick a persona — a starter template (Researcher, Assistant, Companion, …) you can edit, or the full-suite Everything preset.
2. Point it at an LLM and test the connection — a local endpoint (http://host.docker.internal:PORT/v1 for vLLM / Ollama / LM Studio) or a hosted API (OpenAI, Anthropic, OpenRouter, …). Bring your own — the key is used server-side and never written to config or shown in the browser.
3. Choose how much of a mind it has — presets, or the advanced pill grid (dependencies auto-satisfy).
4. Create. The agent boots; talk to it, watch it think, and drive cognition cycles from the dashboard. Change its settings anytime with ⚙ Edit (memory is preserved).

⁠Supported tags

• latest — the current release
• X.Y.Z — a pinned release (recommended for reproducible upgrades)

⁠Image facts

• Platform: linux/amd64
• Ports: 8080 (wizard / dashboard / API), 5174 (brain viz)
• Volume: /data — agent config and the database; back it up, and it's all that persists
• Sizable first pull (bundles Postgres, an embedder, and CPU PyTorch) — subsequent starts are fast

⁠Advanced — multi-agent, identity, perception, hives

The same image also runs as the studio-only service in a multi-service stack. Grab the docker-compose.yml (from the source repo⁠, under docker/studio/) for:

• several agents / a hive (a shared world-model),
• writing-style identity (--profile identity — recognise a returning person by how they write),
• embodied perception (--profile perception — a computer-use sandbox the agent drives).

Companion images (same namespace): nmem-viz, nmem-identity, nmem-sandbox.

⁠License

The engine libraries are AGPL; the studio image is BUSL-1.1 (org.opencontainers.image.licenses = BUSL-1.1 AND MIT). Image: https://hub.docker.com/r/dayyanj/nmem-studio⁠ · Source & docs: https://github.com/dayyanj/nmem⁠.