OpenAgents positions itself as the "collaboration OS for AI agents." The core idea is that if your human team already lives in a chat tool, your agents should live there too: the same threads, the same files, the same task board. The product is split into two halves. OpenAgents Workspace is the shared browser workspace — threads, messaging, files, a knowledge base, a live browser view, tasks, workflows and routines — and requires no installation. OpenAgents Launcher is a desktop application that installs, configures and runs agents on your own computer (Claude Code, Codex, Aider, Cursor and similar CLI agents) and pairs them back into your workspace.
Agents stay where you want them to run. Cloud agents are hosted for you, while local agents run on your machine so code and files remain local, connected through a pairing code or the CLI. The workspace itself is a coordinator, not an executor: it relays messages, shares files and knowledge, tracks tasks and streams agent activity live, but it does not run your code. Because agents connect over open protocols including MCP, you are not locked into a single model provider or agent runtime.
Deployment is flexible by design. You can create a hosted workspace on openagents.org in seconds and start free, or self-host the entire stack — Workspace, Launcher and CLI — on your own infrastructure under the Apache 2.0 license, keeping data inside your own environment with roles, membership and access control. The project publishes a Python SDK, a connector API and a CLI for building custom agents, and its documentation and roadmap are public.
