labor.fun

A standalone, multi-org framework for multi-channel AI assistants. It runs Claude agents in isolated containers with per-group memory and a shared knowledge base, and speaks Slack, Telegram, Discord, and CLI out of the box.

The framework is org-agnostic. Each organization is a self-contained profile you drop into profiles/<name>/ — its identity, knowledge base, people, and runtime state. Nothing about a specific org is baked into the code.

labor.fun is the framework formerly developed as "Breadbrich Engels / NanoClaw". nanoclaw remains the internal codename for the container/agent protocol.

How it's organized

labor.fun/                  ← the framework (org-agnostic, reusable)
├── src/                    Orchestrator: message loop, channels, DB, IPC, scheduler
├── container/              Agent container image + runtime skills
├── kb-ui/                  Admin dashboard
├── rules/                  Core operating rules the agent follows
├── setup/                  Install wizard steps
└── profiles/              ← org instances (only `example` is tracked)
    └── example/            A copy-me template for new orgs
        ├── profile.config.json   identity & config (single source of truth)
        ├── groups/               per-group memory + KB context
        ├── deploy.config         per-org host: paths, service names, user
        ├── container-skills/      optional org-specific agent skills
        └── plugins/               optional org-specific channels & flows

A real org (e.g. profiles/acme/) is created locally from example and is gitignored — its config, KB, plugins, infra, store/, and data/ live only on that org's clone/host, never in the shared framework repo.

Active profile is selected at startup: LABOR_PROFILE=<name> (in .env), else the single profile present, else the repo root (legacy/dev layout).

Quick start for a new org

See docs/NEW-ORG-GUIDE.md for the full walkthrough. In short:

cp -r profiles/example profiles/acme        # 1. copy the template
$EDITOR profiles/acme/profile.config.json    # 2. set identity (name, github org, …)
echo "LABOR_PROFILE=acme" >> .env            # 3. activate it
npm run setup                                # 4. run the install wizard

Extending it

Five consistent extension points — all self-register the same way (see docs/PLUGINS.md):

Extension	Mechanism	Lives in
Profile plugin (org channels & flows)	export default register(api), auto-loaded	<profile>/plugins/
Channel (built-in)	registerChannel() + barrel import	src/channels/
Flow (built-in background integration)	registerIntegration() + barrel import	src/integrations/
Container skill	drop a SKILL.md folder	container/skills/ or <profile>/container-skills/
Infra / deploy	per-org deploy.config	<profile>/deploy.config
Setup step	add to the STEPS registry	setup/
Rules / KB	markdown	rules/ (core) + <profile>/groups/ (org)

Architecture

Slack / Telegram / Discord / CLI
        │
        ▼
   ┌──────────────┐
   │ Orchestrator │  Node.js process — poll → trigger check → identity resolution
   └──────┬───────┘
          │
   ┌──────▼───────┐
   │  Container   │  Docker, isolated filesystem, Claude SDK + MCP tools + skills
   └──────┬───────┘
          │
   ┌──────▼───────┐
   │  IPC Watcher │  Outbound messages, task ops, cross-channel send, KB writes
   └──────────────┘

Database

SQLite at <profile>/store/messages.db. Full schema in schema/tables.md. Core: chats, messages, registered_groups, sessions, router_state. Operational: scheduled_tasks, task_run_logs. Identity: user_identities.

Development

npm install
npm run dev          # Run with hot reload (uses the active profile)
npm run build        # Compile TypeScript
npm test             # Run test suite
npm run typecheck    # Type-check only
./container/build.sh # Rebuild agent container image

License

MIT