Allows receiving and answering questions via Slack, enabling interaction with the fact store through Slack messages.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@goat tell me about my habits from last month"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
A personal assistant backend. It stores facts in Postgres or SQLite, fills them from importers on a schedule, and answers questions about them over Slack, MCP, SMS, or a terminal REPL.
Facts carry two timestamps: when the fact is about, and when goat learned it. Re-running an importer with unchanged data writes nothing.
uv add "goat-butler @ git+https://github.com/phin-tech/goat"Extras: postgres, agent (pydantic-ai), cli, mcp, slack, sms,
weather, migrations, and all for every one of them. SQLite needs none.
uv add "goat-butler[all] @ git+https://github.com/phin-tech/goat"To work on goat itself:
git clone https://github.com/phin-tech/goat
cd goat
uv sync
uv run pytestThe suite runs against both SQLite and Postgres. Postgres comes from
docker compose up -d.
Related MCP server: Cairn
A shed is one deployment: a directory with a shed.py that builds a Shed, and
a .env holding the credentials. goat init writes both.
uv run goat init ~/my-shed
cd ~/my-shed
uv sync
goat migrate
goat ingest
goat replDefaults to SQLite, so this runs with no database server and no API key. Set a
model provider key in .env for goat brief and the REPL's ask.
.env can hold 1Password references instead of secrets. goat resolves them at
startup via the op CLI, so the file is safe to read aloud or paste:
OPENROUTER_API_KEY=op://Private/OpenRouter/credential
| scaffold a new shed |
| create or update the schema |
| run every importer |
| compose a brief; |
| inspect the store |
| interactive session |
| serve the store over MCP on stdio |
goat holds the mechanism: the store, importers, delivery, tools, the CLI. A shed
holds the values: coordinates, model, credentials, prompt, schedule.
OpenMeteoImporter lives in goat because a latitude is a constructor argument.
The latitude lives in your shed.
docs/design/fact-store.md — schema, decisions, and the ones that were reversed
docs/mcp.md — MCP server and Claude Desktop config
docs/slack.md — Slack app setup
docs/weather.md — weather sources
skills/ — Claude Code skills for setup and daily use
This server cannot be deployed
Hosted agent memory: store, search, and recall facts across sessions from any MCP client.
Persistent knowledge graph for AI agents. Remember, recall, and forget facts.
Memory for AI agents that knows what is still true. Typed bi-temporal facts, EU-hosted.
Hosted long-term memory for AI agents via MCP. Recalld updates facts as information changes and retrieves relevant context. Connect with a Recalld EU account.