nightly-labs/openbot

OpenBot is worth a look if you routinely run several agent sessions side by side and want them in one window rather than a stack of terminals. It is a development preview, not a settlement layer for unattended automations.

The problem

You want one agent working through a refactor while another reviews a pull request, a third answers questions about logs, and you want to hand files between them without retyping paths and pasting context. The usual workflow means opening several terminal tabs, losing track of which session belongs to which task, and manually re-pasting progress between them. OpenBot replaces that with a desktop app where each agent has its own workspace and conversation, and work is handed to it through a local queue. The result is a single board of named agents you can review, pause or cancel instead of a pile of shell windows.

What's different

OpenBot's mechanism is one persistent session per agent, backed by a provider CLI you already use: the local Codex App Server and Claude Code are supported directly, while Grok CLI, OpenCode, Gemini, Cursor CLI and Cline connect through ACP. Custom OpenAI-compatible endpoints and custom ACP agents are also accepted, and the app can detect local servers such as Ollama and LM Studio. That breadth is the practical benefit: you can compare agents on the same task inside one interface rather than learning a separate launcher for each.

Around those sessions it adds a FIFO queue with pause, resume and cancellation that survives crashes, per-agent context monitoring that compacts long threads before the model window fills, file transfers, and an embedded browser. Agents can also message each other, which is the piece that distinguishes it from simply opening multiple terminals. The trade-off is that a persistent, self-directed workspace is only as orderly as your task design. The README is explicit that agents run with full local access and no per-action approval after first-launch consent, so the convenience of unattended queues is also the main thing to think through before handing over real work.

Who should try it

A good fit is a developer who already pays for or runs two or more agent CLIs and wants a shared workspace, a visible task queue and lightweight agent-to-agent handoffs on a single machine. The self-hosted server path also suits someone who wants that workspace reachable from a phone or browser, provided they are comfortable running a systemd service or Docker container.

A poor fit is anyone looking for a sandboxed or approval-gated environment, since the documented model grants full access by default, and anyone who needs a commercially licensed tool, because the current license is noncommercial. Teams expecting a vetted, production-hardened collaboration platform should wait; the project labels itself a development preview.

Requirements and limitations

The desktop app targets macOS 13+, Windows 10+ x64, and x64 or arm64 Linux as an AppImage. On Ubuntu 23.10+ and Debian 13, unprivileged user namespaces are restricted by AppArmor and the app exits at launch until the supplied AppArmor profile is installed; the Nix build on Linux also omits the remote desktop runtime. Running it as an always-on server means installing from a terminal, signing in with an email code, and choosing between systemd and a Docker container with a seccomp profile. Expect to manage provider CLI updates as part of ongoing upkeep: OpenBot checks hourly and installs the latest upstream release into its managed copy.

The bigger boundary is the security posture set out in the security notes. Full access, network use and browser control without per-action confirmation mean backups and trusted tasks are your responsibility, not the app's. Licensing is the second boundary: PolyForm Noncommercial 1.0.0 for current versions, with commercial use requiring a separate agreement, though releases up to 0.1.11 remain Apache-2.0.

HotRepo's take

The only recorded signal is a single daily-board appearance on 2026-10-25 with 903 stars and +81 that day; there is not enough comparable history to judge persistence. For a 73-day-old repository, the more useful check is the security section and self-hosted server guide, which tell you whether the default full-access posture and the Linux startup requirements match your setup.