ManyChat’s open-source shadow, built for self-hosted AI agents

What it does This is a self-hosted marketing automation platform that lets teams build drag-and-drop chatbot flows across WhatsApp, Instagram, Telegram, and half a dozen other channels. It bundles a visual flow builder, live chat inbox, contact CRM, broadcast sequencing, and team permissions into a single TypeScript monorepo. An AI agent layer can hand off to OpenAI, DeepSeek, or Gemini, while human operators jump into conversations through a real-time inbox.

The interesting bit The architecture is surprisingly thorough for a 503-star project: a Next.js 16 builder UI, RustFS / S3-compatible object storage, PartyKit real-time sockets, and BullMQ workers split across chat, AI, triggers, and analytics. It also ships an MCP server and a CLI, which suggests the authors are betting on agentic workflows rather than simple no-code integrations. The AGPLv3 license covers the community edition, but enterprise features sit behind a separate commercial license in the same repo—so check the license before you fork for a product.

Key highlights

• Omnichannel by default: WhatsApp, Messenger, Instagram, Telegram, Zalo, TikTok, email, and webchat integrations are first-class citizens, not afterthought plugins.
• AI agent nodes in the visual flow builder can generate responses or trigger actions via OpenAI, DeepSeek, or Gemini.
• Real-time human takeover via PartyKit/PartySocket, plus segmentation, A/B testing, and drip-campaign sequencing for the marketing purists.
• The stack is modern but heavy: Node.js 24, Next.js 16, React 19, Drizzle ORM with pgvector, Redis, and RustFS/S3-compatible storage.
• Includes an MCP server and CLI for users who want to script agent workflows instead of clicking through flows.

Caveats

• The tagline name-checks “OpenClaw” and “Hermes,” but the feature list only mentions OpenAI, DeepSeek, and Gemini, so it is unclear which models are actually supported.
• Enterprise features live in the same repo under a separate commercial license; self-hosters must sort out which code is AGPL and which is not.
• With 503 stars, the project is still building traction against incumbents like ManyChat.

Verdict Worth a look if you need a self-hosted, AI-aware chat marketing stack and do not mind navigating a dual-license repo. Skip it if you are looking for a mature, fully managed SaaS with a support hotline.