Frequently asked questions

A desktop app where you build a team of AI agents, a lead plus specialists, and hand them work in a project chat. The lead splits your request into tasks, assigns them and reviews the results. You watch the work in a pixel-art office, on a Kanban board and in each agent's terminal. See Introduction.

No. Ready-made team templates cover research, marketing, product and e-commerce work, and the Demo provider lets you try everything without an API key. Some tasks, such as editing files or browsing the web, only work with certain providers. Until an installer is published, starting the app takes a few commands (see Install).

The office has 10 worker desks plus a desk for the lead. In a project with more than 10 workers, agents share desks. Agents are shared between projects, and each one works on one task at a time.

English and Turkish. Choose one in Settings → General → Language; the default follows your operating system. Agents answer in the language you write in. Community translations for more languages are welcome.

Partly. Agentbits has no server and no account, and Demo doesn't call any AI model. Real work needs your AI provider, which is an online service, so it needs an internet connection. Local models aren't supported yet.

Yes, for personal and other noncommercial use. Charities, schools, public research and government organizations may use it too. Every feature is available to everyone. Your AI provider may charge for model usage, which is separate from Agentbits.

Not under the standard license, which covers noncommercial use only. For commercial use or any other permission, get in touch through GitHub issues.

No. It is source-available under the PolyForm Strict License 1.0.0. The code is public on GitHub, but you may not modify it, make derivative works or redistribute it.

Yes. The contributing guide gives you permission to change the code only to prepare a pull request, and explains the terms. Please open an issue before a larger change so the approach can be agreed first.

Use the Buy me a coffee button in the status bar, or visit buymeacoffee.com/godilman. Support is entirely voluntary and nothing is locked behind it. Bug reports, ideas and translations help a lot too.

Claude Code CLI, Codex CLI, Anthropic API, OpenAI API and Demo (simulation). You set a default and can pick a different provider and model per agent. Gemini and local models are not supported yet. See Providers.

Not to try it: Demo needs nothing. Claude Code CLI and Codex CLI use their own sign-in on your computer (claude auth login, codex login), so Agentbits needs no key for them. Only the Anthropic API and OpenAI API providers need a key.

Yes, through Claude Code CLI: install it, sign in with claude auth login, and Agentbits runs it with your own account. Agentbits does not offer signing in with a Claude account inside the app.

Agentbits itself costs nothing. Model usage is billed by your provider: per token for an API key, or through your account for a CLI. Agentbits shows token usage per task, agent and project in agent terminals, the status bar and Reports → Token usage. A cost in dollars appears only when Claude Code reports it.

Not yet. There is no cap on tokens, time or money per task or project; it is on the roadmap. Until then, keep Require approval on for agents you want to control, watch token usage, and use Pause all to stop everything.

Only agents on Claude Code CLI or Codex CLI with the Network permission (Codex also needs Write files).Anthropic API and OpenAI API agents can't browse. The built-in Browser tab is for you; agents can't use it yet.

Your data is stored on your computer. What agents work with (your messages, their prompts and the file contents they may read) is sent to the AI provider you chose. Other traffic goes only to the connections and apps you set up, to GitHub when you install a skill, and to the sites you open in the Browser tab.

No. There is no telemetry, no analytics, no account and no Agentbits server. The Feedback button only opens a prepared email in your own mail app.

In a local SQLite database: %APPDATA%\Agentbits on Windows, ~/Library/Application Support/Agentbits on macOS and ~/.config/Agentbits on Linux. Backups, restore and reset are in Settings → Data management. See Settings.

Claude Code and Codex CLI agents with Write files really change files in the project folder, and an agent with Terminal can run any command with your user's rights. Use an empty folder or a git branch, keep Require approval on, and give Terminal only to agents you trust. API agents don't touch the project folder, and deleting a project never deletes its folder. See Permissions.

They are encrypted with your operating system's keychain (Windows DPAPI, macOS Keychain, Linux Secret Service) and are never sent to the interface, prompts, terminals or logs. If no keychain is available, Agentbits refuses to store secrets. See Security.

Privately, not in a public issue: use Report a vulnerability on GitHub. Include the Agentbits version, your operating system, the provider and the steps to reproduce, but no real keys or personal data.

Not yet. Windows packages, and later macOS, will be published on GitHub Releases once a release has been built and checked. Until then you run Agentbits from source with Node.js 22 or newer. See Install.

Windows is the main target. Running from source works on Windows, macOS and Linux. The macOS package hasn't been tested on a real Mac yet, and there is no Linux package.

The installers are not code-signed yet, so Windows SmartScreen warns you. Click More info, then Run anyway.

No. Install a newer version over the old one; your data folder is kept. When running from source, pull the latest code and run npm ci.

Open an issue on GitHub with your operating system, the Agentbits version (shown in Settings → General), the provider and the steps to reproduce. For ideas, describe the problem you want to solve. You can also use the Feedback button in the status bar.

Yes. A new language is a folder of translation files plus a few lines of setup; the steps are in the translation guide. Open an issue first so work isn't duplicated.