litellm-labs/litellm-agent-control-plane

LiteLLM Agent Control Plane gives your team a single dashboard and API to create, manage, and run AI agents across multiple different agent platforms. Instead of jumping between separate consoles and tools, everyone works from one shared interface. The core idea is that different AI agent runtimes (Claude Managed Agents, Cursor, DeepAgents, Hermes, OpenCode, and OpenClaw) each have their own way of doing things. This project sits on top of all of them and translates your instructions into whatever format the underlying runtime needs. You create an agent through the unified dashboard, pick which tools and skills it should have access to, choose which runtime it runs on, and fire it off. The platform also handles session management so agents keep their context between runs, scheduled agent execution, and persistent memory so agents remember things across sessions. This is built for teams where multiple people need to create or use AI agents. A concrete example: a product manager or developer can build an agent, hand it to the rest of the team, and they can all use it without needing direct access to the underlying provider accounts. Nobody on the team needs to log into Anthropic's console or configure credentials themselves, the access is managed centrally through the control plane. Getting started requires Docker Desktop and a single command to spin up the web interface, a database, and a default runtime. From there you can add other runtimes as needed. The README doesn't go into deep detail on how each runtime integration works or what the tradeoffs are between them, but the pitch is straightforward: one place to call all your agents, one set of access controls, and one UI your team shares.