Baba

A Slack-based monitoring teammate for your service. You talk to it, it watches your stuff, it pages you when things go sideways, and it remembers what it learns so every conversation makes it sharper.

"Mongo is the ledger. The graph is the understanding. Chat is the interface."

What it is

Baba is a chat-driven service monitoring agent. Instead of YAML dashboards and alert rules, you describe your service to it in Slack — what endpoints matter, what "healthy" looks like, what you've learned about tricky corners. It monitors on a schedule, posts alerts when something looks wrong, and engages you in conversation to get sharper over time.

Unlike a typical bot, Baba has memory. It builds a living graph of what it knows about your service — endpoints, recurring incident patterns, remembered facts about you, domain entities — and stale beliefs age out on their own.

The three pillars

• Ledger — append-only operational truth in Mongo: runs, incidents, tool traces, metrics, alerts outbox, LLM cost timeseries.
• Graph — what Baba thinks. Typed nodes + edges (also in Mongo). Every claim cites the run that produced it.
• Chat — every teaching, correcting, or approving interaction happens in Slack. There's no admin web UI. Chat is the authoring surface, full stop.

Architecture in one diagram

flowchart LR
    User["Operator (Slack)"]
    Slack[Slack]
    Bot["Bot process<br/>(chat + alerts)"]
    Worker["Worker process<br/>(ticks + nightly)"]
    Mongo[(MongoDB<br/>ledger + graph)]
    LLM["LLM<br/>(Claude / OpenAI-compat)"]
    Service["Monitored service<br/>(HTTP + MCP + Mongo)"]

    User <-->|mentions, reactions, /baba| Slack
    Slack <-->|Bolt events| Bot
    Bot <-->|read/write| Mongo
    Worker -->|read/write| Mongo
    Worker -->|probe| Service
    Bot <-->|chat, synthesis| LLM
    Worker <-->|tick, digest, review| LLM

Two processes (worker + bot) sharing one Mongo database. They don't talk to each other directly — state flows through alerts_outbox, scheduled_tasks, and conversations.

Quickstart

pnpm install
cp babusya.config.example.ts babusya.config.ts   # edit: service, slack channels, sources, auth
pnpm init                                   # create Mongo collections + indexes + TTLs
pnpm validate                               # config dry-run with auth check per source
pnpm worker        # terminal 1
pnpm bot           # terminal 2

You'll need MONGO_URL, ANTHROPIC_API_KEY, SLACK_BOT_TOKEN, SLACK_APP_TOKEN (or SLACK_SIGNING_SECRET), plus whatever env vars your sources declare.

Full instructions: docs/setup.md.

Documentation

• docs/product.md — what Baba is and how it thinks
• docs/setup.md — first-time install (Slack app, config, env, run)
• docs/configuration.md — every babusya.config.ts field with examples
• docs/interactions.md — chat patterns, reactions, /baba commands, social beats
• docs/mongo-backup-restore.md — backup procedures and restore drill
• idea.md — canonical product spec (~1400 lines)
• HARNESS.md, RUNBOOK.md, COLLECTIONS.md, AGENTS.md — operations + internals

Stack

Node 20+, TypeScript, run under tsx (no build step). MongoDB standalone (no replica set). Anthropic + OpenAI-compatible LLM adapters. @slack/bolt for Slack. @modelcontextprotocol/sdk for MCP servers. pino for structured logs.

Contributing

Read CLAUDE.md for the working agreement and load-bearing invariants. Read idea.md for the product spec. Changes to product behavior must update idea.md in the same session.

License

ISC.