Description
Codex in one terminal, Claude Code in another, Gemini in a third, each doing its own thing. Who is working on what, which one is stuck, and how one result reaches the next all depend on you switching windows and copy-pasting.
OpenBot brings them into one desktop team of AI agents. Every agent has its own name, instructions and workspace, and they message each other, hand off tasks and share files. One says "the sign-in test fails on Safari, Linus, can you take it?", another fixes it and asks a third to review. You only get pulled in when someone needs a decision, and the question shows up right in the app as choices to click.
The practical part: it uses the plans you already pay for. ChatGPT, Claude, Gemini and Grok plans plug in directly, and so do OpenAI-compatible endpoints, Ollama and LM Studio, so you don't buy a separate AI for it. Workspaces, conversations, files and browser data stay on your own computer.
A team of agents: Codex, Claude Code, Gemini, Grok, OpenCode, Cursor, Cline and more, plus custom agents through ACP.
Agents that collaborate: they message, reply, and send images and files; in a channel one agent owns the task and can delegate, pause or reassign it.
A persistent embedded browser: agents open pages, inspect them and drive them, with sign-ins kept between sessions.
Queues that survive crashes: first in, first out, with pause, resume and cancel.
Automatic context compaction: each agent's context is monitored and compacted before a long thread fills the window.
Many platforms: desktop apps for macOS, Windows and Linux, with iOS and Android apps too.
Know the risk first: it is a development preview, and agents run with full file and command access without per-action confirmation, so run only tasks you trust and keep backups. Free for personal non-commercial use under the PolyForm Noncommercial license.
OpenBot brings them into one desktop team of AI agents. Every agent has its own name, instructions and workspace, and they message each other, hand off tasks and share files. One says "the sign-in test fails on Safari, Linus, can you take it?", another fixes it and asks a third to review. You only get pulled in when someone needs a decision, and the question shows up right in the app as choices to click.
The practical part: it uses the plans you already pay for. ChatGPT, Claude, Gemini and Grok plans plug in directly, and so do OpenAI-compatible endpoints, Ollama and LM Studio, so you don't buy a separate AI for it. Workspaces, conversations, files and browser data stay on your own computer.
Features
A team of agents: Codex, Claude Code, Gemini, Grok, OpenCode, Cursor, Cline and more, plus custom agents through ACP.
Agents that collaborate: they message, reply, and send images and files; in a channel one agent owns the task and can delegate, pause or reassign it.
A persistent embedded browser: agents open pages, inspect them and drive them, with sign-ins kept between sessions.
Queues that survive crashes: first in, first out, with pause, resume and cancel.
Automatic context compaction: each agent's context is monitored and compacted before a long thread fills the window.
Many platforms: desktop apps for macOS, Windows and Linux, with iOS and Android apps too.
Know the risk first: it is a development preview, and agents run with full file and command access without per-action confirmation, so run only tasks you trust and keep backups. Free for personal non-commercial use under the PolyForm Noncommercial license.
