Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseCode / SDK · ActionableGitHub Trending AI→ original

OpenBot Brings Codex, Claude and Grok Into One App: 794 GitHub Stars in a Day

If you run Codex, Claude and Grok in separate terminals, OpenBot brings them together into one local app. Each agent gets its own working folder and its own conversation, plus a message queue, file exchange and a built-in browser. The nightly-labs project picked up 794 GitHub stars in a single day and connects more than a dozen providers through the ACP protocol. Run it only on trusted tasks and keep backups, because agent actions are not confirmed one by one.

AI-processed from GitHub Trending AI; edited by Hamidun News
OpenBot Brings Codex, Claude and Grok Into One App: 794 GitHub Stars in a Day
Source: GitHub Trending AI. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceGitHub Trending AI ↗
Editor-in-Chief
Key metric #1Works with Codex, Claude, Grok, Gemini, Cursor, Cline, Devin, Copilot
Key metric #2The server can run on a VPS or in Docker, and there is an iPhone app
The license is not free softwareinstalling via Nix requires allowUnfree

There are plenty of coding agents now, and each one has its own terminal, its own login and its own history. One lives in Codex, another in Claude Code, a third in Grok CLI. Switching between them is awkward, and getting them to work together is harder still. The OpenBot project from nightly-labs tries to solve exactly this: it gathers different agents into one desktop app. In a single day, the repository gained 794 stars on GitHub Trending.

What's inside: a team of agents instead of a scatter of terminals

OpenBot calls itself a local workspace for persistent AI colleagues. Each agent gets its own working folder and its own conversation. Picture a small team: one agent writes tests, a second digs through logs, a third reads documentation in the browser. Each has its own desk and its own context, so they don't get in each other's way.

Under the hood, it connects sessions for Codex, Claude, Grok, OpenCode, Gemini, Cursor, Cline, Kimi, Devin, GitHub Copilot, Mistral and Hermes. For anything else there are custom ACP agents and any OpenAI-compatible endpoint. The app can detect local model servers such as Ollama and LM Studio. For each provider, OpenBot downloads and pins the required CLI version, but if you already have your own installed, it uses that instead.

Some useful extras:

  • FIFO message queues with pause, cancel and persistence after a crash;
  • per-agent context monitoring and automatic compression, so a long conversation doesn't hit the model's window limit;
  • exchange of messages, images and files between agents;
  • shared channels where a task has a single owner, and it can be handed to another agent, stopped or restarted;
  • a persistent built-in browser that agents open and control themselves;
  • an optional Computer Use mode on macOS, Windows and Linux.

There is also a skills marketplace. A skill is described by a folder following the Codex structure, and you can ask an agent to create a skill itself from a repeated process. Agents from Grok Bot can be imported from a file.

Where it runs

Desktop builds are available for macOS 13+ (Apple silicon and Intel), Windows 10+ on x64, and Linux (x64 and arm64) as an AppImage. For always-on use, you can run OpenBot as a server on a VPS or a home server with Ubuntu 24.04 or 26.04 and systemd. The same server is packaged as a Docker image with a seccomp profile. You can connect to it from the desktop app or from the iPhone app. Optional accounts work through one-time codes sent by email, and the accounts API runs on Cloudflare Workers and D1.

The authors stress that "local" does not mean "offline." Codex still talks to OpenAI, Claude to Anthropic, Grok to xAI, and so on. OpenBot does not copy your provider credentials; it uses the login that already exists in the local CLIs.

Limitations and risks

The main warning sits right in the description. After consent on first launch, agents read and modify files, run commands, access the network and control the browser without confirming each action. The authors advise running only trusted agents and tasks, keeping backups and checking results. For a work laptop holding important data, that is a strong argument for a separate virtual machine or container.

There are other caveats:

  • OpenBot is not free software, so in Nix you have to permit unfree packages via allowUnfreePredicate;
  • on Ubuntu 23.10+ and Debian 13 the app will not start until you install an AppArmor profile;
  • remote desktop on Linux works only in an X11 session, not under Wayland;
  • Gemini sign-in requires a graphical session on the same host, so it can't be completed on a headless server;
  • for some providers, such as Cursor, Cline and Devin, agents stay on your computer and are not visible to team members.

What this means in practice

If you already pay for several agent subscriptions, OpenBot saves time on switching and lets you compare agents on the same task in one shell. For teams, the appeal is running a shared agent server and connecting to it from a phone. But the project is young: releases come out almost daily, so the interface and behavior may change quickly. Start in an isolated environment with a few non-critical tasks, and only then hand agents your working repositories.

Story

Key Questions Answered

What's inside: a team of agents instead of a scatter of terminals?

OpenBot calls itself a local workspace for persistent AI colleagues. Each agent gets its own working folder and its own conversation. Picture a small team: one agent writes tests, a second digs through logs, a third reads documentation in the browser. Each has its own desk and its own context, so they don't get in each other's way. Under the hood, it connects sessions for Codex, Claude, Grok, OpenCode, Gemini, Cursor, Cline, Kimi, Devin, GitHub Copilot, Mistral and Hermes.

Where it runs?

Desktop builds are available for macOS 13+ (Apple silicon and Intel), Windows 10+ on x64, and Linux (x64 and arm64) as an AppImage. For always-on use, you can run OpenBot as a server on a VPS or a home server with Ubuntu 24.04 or 26.04 and systemd. The same server is packaged as a Docker image with a seccomp profile. You can connect to it from the desktop app or from the iPhone app. Optional accounts work through one-time codes sent by email, and the accounts API runs on Cloudflare Workers and D1.

What are the key details on Limitations and risks?

The main warning sits right in the description. After consent on first launch, agents read and modify files, run commands, access the network and control the browser without confirming each action. The authors advise running only trusted agents and tasks, keeping backups and checking results. For a work laptop holding important data, that is a strong argument for a separate virtual machine or container.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Hands-On AI Engineering · Hamidun Academy

Build & Ship Production AI Agents Without Boilerplate Code

Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.

Start 72h Free Trial →Vibe-Coding Intensive
What do you think?
Loading comments…