GridBear

GridBear is a secure, open-source platform to run AI agents in production. Wire multiple LLM runners (Claude, OpenAI, Gemini, Ollama) to multiple channels (Telegram, Discord, WhatsApp, web chat) with a shared plugin ecosystem for tools, memory, workflows, and per-user identity.

What's in the box

• Multi-runner — switch runner per-agent; hot-reload config from the admin UI
• Multi-channel — receive messages from chat apps, reply through the same agent pool
• Plugin system — 40+ plugins with manifest.json discovery, dependency resolution, hot-reload
• MCP gateway — SSE-based gateway with per-user OAuth2 connections, circuit breakers, rate limiting
• Memory — episodic and declarative memory on PostgreSQL + pgvector
• Workflow engine — visual DAG editor with agent, tool, condition, transform, and approval steps
• Admin UI — web-based management; every plugin can ship its own admin page
• User portal — per-user dashboard, service connections (Gmail, Google Workspace, …), MCP tool toggles

Get started

• Getting Started — clone, configure, boot the stack
• Architecture — containers, plugin system, message flow, data model
• Plugin Development — build your own channel / service / runner / MCP plugin
• API Reference — REST endpoints, authentication, rate limits

Community

• Discord — live chat and Q&A
• GitHub Discussions — RFCs, Ideas, Q&A
• GitHub Issues — bug reports, concrete feature requests

Early-stage project

GridBear is currently pre-1.0. We don't recommend production deployments yet — see the warning on the GitHub README for details. Feedback and contributions are very welcome.