OpenBot runs Codex, Claude, Grok, and other AI models in a single desktop app with task queues, file sharing, and agent-to-agent messaging. Models connect to your endpoints or cloud services.

OpenBot is a local-first desktop workspace that lets you run multiple AI models—Codex, Claude, Grok, Gemini, Cursor, and Cline—side-by-side in one app. Each agent gets its own persistent workspace, conversation history, and dedicated task queue. The README describes it as a workspace for “persistent AI teammates” with 3,381 GitHub stars as of late October 2026.

What it does

Instead of switching between browser tabs and chat windows, OpenBot consolidates AI agents into a single interface. Each agent runs in its own environment with independent model sessions and local workspaces. You can create agents by editing their instructions and avatar directly in the app on desktop or mobile, with section review before saving, then reuse them.

The app includes FIFO message queues for each agent, letting you queue tasks, pause or resume them, and cancel in-progress work. The README notes that queues are crash-safe and persistent.

Local-first architecture and endpoints

OpenBot is local-first but not offline-only. The README states that Codex connects to OpenAI, Claude to Anthropic, Grok to xAI, and Gemini to Google. You can also point agents at custom OpenAI-compatible endpoints or local model servers like Ollama and LM Studio. The README notes that OpenBot detects installed local servers and agents automatically.

Per-agent model settings, reasoning mode, profile, notification state, and browser state are stored locally. Diagnostics exports are privacy-safe and available from the account menu.

Agent collaboration and browser control

Agents can message each other, reply to messages, and transfer files between workspaces. A persistent embedded browser lets agents open pages, inspect content, and control navigation.

Optional Computer Use is available on macOS, Windows, and Linux through a cua-driver binary included in releases. OpenBot starts this as a child process and provides it to each provider that supports it.

Security model and who it is for

OpenBot is a development preview. The README warns that agents currently run with danger-full-access and approvalPolicy: never. After you grant first-launch consent, agents can read and modify files, run commands, use the network, and control the browser without per-action confirmations. The authors recommend running only agents and tasks you trust, keeping backups, and reviewing the security section before use.

Installation is available for macOS 13 or newer (Apple silicon or Intel), Windows 10 or newer (x64), and Linux (x64 or arm64 AppImage). On Ubuntu 23.10 and Debian 13, an AppArmor profile must be installed to allow unprivileged user namespaces; the profile ships inside the AppImage at resources/linux/openbot.apparmor.

Go deeper

Get the latest AI updates in the Tech AI Magazine app
iPhone · Android