// the find
crestalnetwork/intentkit
IntentKit is an open-source, self-hosted cloud agent cluster that manages a collaborative team of AI agents for you.
IntentKit is a self-hosted platform for running teams of AI agents that can call each other, post to social media, and execute Web3 transactions. It targets developers who want a cloud-native alternative to local LLM setups, with a clear bent toward crypto and social media automation use cases.
The multi-runtime split is sensible: Python for the agent core, Go for the Telegram and WeChat integrations (which handle rate limiting, outage tracking, and SSE properly), and Next.js for the frontend — each built independently with its own tooling. The MCP client support in intentkit/clients/mcp/ plugs into the growing Model Context Protocol ecosystem without you having to wire it yourself. Autonomous scheduling is first-class, not bolted on — agents can run on schedules rather than only reacting to messages. OpenTelemetry tracing is baked in via intentkit/config/tracing.py, which matters when you're debugging why agent B called agent C at 3am.
They explicitly don't accept pull requests — 'due to the rapid pace of AI development' — which makes 'open source' mostly mean 'you can read it', not 'you can shape it'. The agent_docs/todo/ directory is committed to the public repo and contains internal billing and pricing plan files (team-billing-p0, pricing-plans-p1), which tells you where this is heading: a SaaS product with an open-source front door. Documentation is hosted at an external domain (intentcat.com) rather than in-repo, so when they rebrand or pivot you lose it. The operational footprint for self-hosting is non-trivial — Docker Swarm, Redis, Postgres, S3, and a separate Go process for each messaging platform — and since you can't contribute fixes, you're on your own when something breaks.