# Diabase

*Read this in: [Italiano](README.it.md)*

**Agent-safe control plane for Supabase, self-hosted or cloud.**

Diabase lets AI agents (Claude Code, or any LLM via API) operate and manage your entire Supabase backend — schema, edge functions, auth workflows, storage — while you watch every move from a modern GUI and stay in charge of the decisions that matter.

> ⚠️ **Status: early development.** Nothing here is ready for production use yet. Follow along or star the repo — the first usable release will be announced.

## Why

Supabase's official MCP executes directly against your production project: no plan, no diff, no staged approval, no audit trail. Diabase is the opinionated layer that makes agent-driven backend management safe enough for a company to allow:

- **Plan & Approve** — large operations become a readable plan you approve, reject, or amend; destructive steps are highlighted
- **Audit trail** — every action, human or AI, recorded append-only
- **Adapters** — the same experience against Supabase self-hosted or Supabase Cloud
- **Projects** — Claude-style workspaces with their own system prompt, context files and autonomy level
- **Your LLM, your choice** — Claude Code CLI (subscription) or pay-per-use API, configured in settings

## Quickstart (Docker)

```bash
docker compose up -d
```

Open http://localhost:8000 and sign in with `admin` / `admin` — you are forced
to rotate the password on first login. Data (database and generated secret key)
lives in the `diabase-data` volume; set `DIABASE_PORT` or `DIABASE_ALLOWED_HOSTS`
in the environment to change the defaults.

**Optional knowledge graph** — a temporal graph memory (Graphiti + Neo4j) over
audited actions and chat:

```bash
docker compose --profile graph up -d
```

Then enable it in the graph settings with two connections: an extraction LLM
(Anthropic or OpenAI-compatible) and an OpenAI-compatible embedder (Ollama
works). Without the profile — or with Neo4j down — Diabase behaves exactly as
before: the graph is an enhancement, never a requirement.

> The Claude Code CLI backend needs the CLI installed, which the image does not
> ship. Inside the container use API connections (Anthropic or any
> OpenAI-compatible endpoint); the Claude Code backend works in a source checkout.

For development, run from source instead:

```bash
uv run python manage.py migrate
uv run python manage.py runserver 8001
```

## License

[Apache 2.0](LICENSE)