Frequently asked questions

#Is LoCrew open source?

No. LoCrew is source available under the PolyForm Perimeter License 1.0.1. It is not an open-source licence in the OSI sense, because it rules out one kind of use.

• Permitted: any purpose but the one below, including personal use and use inside a company, commercial work included. You may fork, change and redistribute the code, as long as recipients get the licence terms.
• Not permitted: providing others with a product that competes with LoCrew, in any form, even free. A fork you use yourself, or publish to contribute back, does not compete.

The licence does not grant rights to the LoCrew name or logo. The crew portraits and application icon have no established distribution rights, so the licence is not offered for them. This is a summary; the licence text is what applies.

#Does the AI run on my computer?

The app does: LoCrew, its database and your credentials stay on your computer. The AI runs wherever the runtime runs:

• Claude Code and Codex call Anthropic's and OpenAI's services under your own login.
• Hosted providers run the model on their servers.
• A local model server (Ollama, LM Studio, vLLM, llama.cpp) runs the model on your machine.
• External A2A agents run wherever they are hosted.

External services see what you send them. See Local data and privacy.

#Where is my data?

In one folder: a SQLite database (locrew.db) plus an attachments/ folder for images. The path is in Settings → About → Data folder. API keys are encrypted with your operating system's keychain.

#Does it work offline?

The project does not state this. LoCrew needs no server of its own. An agent on a local model server only talks to your machine, so it can work without internet access. Claude Code, Codex, hosted providers, remote MCP servers, A2A agents and HTTP steps need their own network access.

#Is there telemetry?

The project's documentation makes no statement about telemetry, and no analytics code was found when these docs were written.

#What does it cost?

The project does not state a price for LoCrew. You pay for the runtimes you use: your Claude Code or Codex plan, or your provider's API. LoCrew shows what it measured next to your provider's own quota, and never merges the two. Default safety limits are $2 per run per agent and $5 per chain, but they only apply to Claude Code and Codex, which report spend. Model agents report tokens only.

#Do I need an API key?

Not if you use Claude Code or Codex (they use your existing CLI login) or a local model server. Hosted providers need their own key.

#Which operating systems are supported?

The README says LoCrew is built and exercised on Windows and macOS, and that Linux is configured but has not been run by hand. Automated tests run on Windows and Linux. Flows have been tested on Windows. See platform status.

#Is there an installer?

Not yet. Run LoCrew from source with npm install and npm run dev. See Installation.

#Which runtimes and providers are supported?

Claude Code, OpenAI Codex (beta), any model from a connected provider, and external agents over Agent2Agent (beta). Provider presets cover OpenAI, Anthropic, Google Gemini, Azure OpenAI, AWS Bedrock (through its OpenAI-compatible endpoint), OpenRouter, Groq, Together AI, Fireworks AI, Hugging Face, Ollama, LM Studio, vLLM, llama.cpp server and any OpenAI-compatible or custom endpoint. Native AWS SigV4, Vertex AI and Azure Entra ID sign-in are not supported. A Gemini CLI runtime is not available.

#Does LoCrew use my Claude Code settings or MCP servers?

No. Claude Code agents run without your personal ~/.claude settings, hooks, MCP servers and CLAUDE.md. Connect MCP servers in LoCrew and select skills per agent.

#Can agents damage my files?

Claude Code and Codex agents can read and change files in their working directory, within their access level. Access levels are limits, not a sandbox: the CLI runs as your user account. Use Read only or Ask first, and a virtual machine or container for real isolation. See Permissions.

#Why didn't my agent answer in a channel?

In a channel, only agents you @mention are woken. A message that mentions nobody wakes nobody. See Channels.

#Can agents talk to each other forever?

No. By default a chain stops after 6 runs (counting the ones your message starts), after 3 consecutive runs of one agent woken by other agents, or after $5 of measured spend, and there is a master switch to turn hand-offs off. See Limits.

#Does "Allow for 1 hour" survive a restart?

No. Work sessions are kept in memory only; closing LoCrew always returns to asking.

#Can I use MCP servers that need OAuth?

Not yet. Remote servers can use a bearer token or a custom header.

#How do I report a security issue?

The project's security policy is a draft and there is no private reporting channel yet. Do not post vulnerabilities in public issues.

#Where do I report a bug?

Open an issue on GitHub with your OS, Node.js version, LoCrew version, the steps to reproduce and any logs with keys removed.