OpenBot is a free desktop application that runs a small team of persistent AI agents on your own machine instead of in a cloud service. It is aimed at developers who already own a coding model subscription or API key (Claude Code, Codex, Gemini and similar) or who run models locally, and who want several agents working in parallel on their own files without paying a per-seat cloud fee.
What it does
OpenBot gives you a roster of agents rather than a single chat window. Each agent has its own name, its own instructions, its own working directory and its own conversation history that survives restarts, so one agent can live in a backend repo while another sits in a docs folder.
The agents are not isolated. According to the README they can message each other, hand work off, and share files between their workspaces, and one of them can drive a browser built into the app. Workspaces, conversations and browser data all stay on the machine, which is the main difference from a hosted teammate product.
The app is distributed for macOS, Windows and Linux, with a Docker option as well.
How it works
OpenBot ships no model of its own. It is a "bring your own provider" shell: you point it at a coding agent or an API you already have access to. The README lists Codex, Claude Code, Grok CLI, Gemini, OpenCode, Cursor CLI, Cline CLI, any OpenAI-compatible endpoint, and fully local models served through Ollama or LM Studio.
Under the hood the project is built with Electron and SolidJS, with Bun and SQLite behind the interface, which is why agent state and conversation history can sit in a local file instead of an account. The repository page shows roughly 2.4k stars, 245 forks and 187 open issues: numbers reported by GitHub itself, not an independent assessment of how finished the app is.
Getting started
The repository is github.com/nightly-labs/openbot and the project's site is openbot.run. Because this is an Electron desktop app, expect to install a prebuilt build for macOS, Windows or Linux, or to run the Docker image, rather than cloning and compiling. The materials do not name an install command, so follow the repository's own instructions.
The money question matters more than the install. OpenBot itself is free, and no OpenBot subscription exists. The intelligence is not free: you either plug in a paid API key or subscription you already hold and pay that provider's usage bills, or you run a local model through Ollama or LM Studio on your own GPU and RAM. The project does not publish hardware requirements, so if local inference is the plan, the practical requirement is whatever the model you pick needs, and several agents working at once will ask more of the machine than a single chat would.
When to use it / when not
The catch is the licence. Newer versions of OpenBot are licensed for noncommercial use only, which means the free app a hobbyist can install may not be something a company can simply roll out to a team, so check the licence in the repository before it reaches a work laptop. The large open issue count is a second signal that this is a young product.
Use it instead of a hosted AI teammate when the work has to stay on your disk, when you want more than one agent with separate memory and directories, or when you already pay for a model and would rather not pay again for a seat. Do not use it if one terminal session of Claude Code or Codex already covers your work, if you need a commercial licence today, or if you have neither a paid API key nor hardware that can run a local model.
Alternatives
The team's own pitch, on Product Hunt, is that OpenBot is a "Grok Bot alternative". Grok Bot is xAI's cloud-hosted persistent AI teammate, and the README places it inside Cursor Ultra at $200 a month or SuperGrok Heavy at $300 a month, with usage billed on top and no spend cap yet. Those are the vendor's published prices as the README reports them, not a comparison anyone ran. The cheaper option most readers already know is running a single coding CLI in a terminal; OpenBot's argument is the team, the shared files and the per-agent memory, not model quality.
One more thing worth knowing: the authors are not an AI lab. Synthetify Labs, the Poland-registered company behind the "nightly-labs" org, was previously known for Nightly, a multichain crypto wallet, so judge the app on what it does rather than on a track record in agent tooling. Take OpenBot seriously if you keep several long-lived agents across different repositories and want their memory, files and browser sessions on hardware you control, since it is a sensible local answer to something otherwise sold as a monthly seat. Be more careful if you are choosing for a business, where the noncommercial licence on recent versions, not the feature list, is what will decide it.