Skip to content

FAQ

No. The site only hosts documentation and installers. Foxwarm runs on your own machine or server; your conversations, files and keys stay there.

Foxwarm is free and open source under the MIT license. You pay your model provider for the tokens you use, or nothing if you run a local model. Add pricing to a model to see an estimated cost for each reply.

Any endpoint that speaks the OpenAI Chat Completions, OpenAI Responses or Anthropic Messages format, including local servers and gateways. You can configure several providers and switch per Session. See Set up your first model and Model options and routing.

How is this different from a coding agent in my terminal or editor?

Section titled “How is this different from a coding agent in my terminal or editor?”

Foxwarm is a server you keep running. Conversations persist and can run for days, survive restarts and are compacted rather than cut off. Work can be handed to other Sessions and Agents, run on other machines through Nodes, and be reached from a browser, a phone or a team chat. See How Foxwarm works.

Yes, on Apple Silicon Macs with macOS 15 or later, using the same installer command as Linux. Intel Macs are not supported as a server, and there is no macOS desktop app. Any Mac can still join another server as a Node. See macOS (Apple Silicon).

At your own installation, using the link printed when it starts, such as http://localhost:3001/#token=.... If you lost it, read state/token in your data directory; see Open the WebUI.

The WebUI token, Node pairing tokens and external MCP identity tokens are different credentials. Use the one for the interface you are connecting to.

Yes. The WebUI works in mobile browsers; if your server is not reachable from the phone, put it behind your usual VPN or reverse proxy. You can also talk to a Session through a chat app with Channels.

Do not share your sign-in token; it is a full administrator credential. Give them an identity that can only open selected Sessions, see Scoped WebUI access, or add them to a Channel’s allow list.

The data directory Foxwarm is using has no state/models.yaml yet. Save a model configuration in Setup → Models. If you expected an existing configuration, check that Foxwarm is pointed at the right data directory before creating a new one.

Models saved fine, but chat fails. What should I check?

Section titled “Models saved fine, but chat fails. What should I check?”

Saving checks the configuration but does not call the provider. Check the provider type, base URL, exact model ID and API key, then send a short text-only message. Turn on hosted tools only after that works. Errors are logged under state/logs/.

Can different Sessions use different models?

Section titled “Can different Sessions use different models?”

Yes. Each Session has its own model and reasoning effort. Change them from the menus next to the message box or with /model. Child Sessions can use a different model from their parent.

What happens when a conversation gets very long?

Section titled “What happens when a conversation gets very long?”

Foxwarm compacts it: older parts are summarized into blocks you can expand, and the original messages stay in the archive where the agent can recall them. You can also run /compact yourself. See Long conversations.

Yes. Sessions, history, queued messages and timers are saved. After a restart, Sessions that were in the middle of a turn are resumed so the agent can check what was interrupted. If one stops instead, /continue resumes it without adding a message.

How do I give an agent lasting instructions?

Section titled “How do I give an agent lasting instructions?”

Put them in the Agent’s memory files under agents/<agent>/memory/, or ask the agent to remember them. See Agents, Sessions and memory.

Use the stop button in the WebUI or send /stop.

Use tool rules: per instance, and per Agent so that helpers an agent creates get less access than it has. See Access control.

No. A command on a Node runs as the user that runs the Node client and can do what that user can do. Use a container or a dedicated user and machine when you want stronger separation, and tool rules to decide which Agents may use which Node.

No. Models, MCP servers and their credentials stay on the Foxwarm server. A Node only runs the file, shell and browser operations it is given.

Why can’t Foxwarm reach a service at localhost?

Section titled “Why can’t Foxwarm reach a service at localhost?”

localhost means the machine or container making the request. Inside a container it is that container; on a Node it is that Node. Use an address reachable from where the request is made.

For a model service on the Docker host, use host.docker.internal; the repository’s Compose file maps it to the host. The service must also listen on an address the container can reach.

A Channel is connected, but my messages get no reply. Why?

Section titled “A Channel is connected, but my messages get no reply. Why?”

Check, in order: the sender’s platform user ID is in allowedUsers (send /whoami to see it), the conversation is attached to the Session you expect, and the bot has permission to post there. The Channels guide walks through it.

QQ Bot groups need an @mention by default; set requireMention: false to handle every message. Group members who are not on the allow list are refused unless you enabled /channel dangerously-allow-all-users yes in that group. See Who can talk to the agent.

Saving Setup → Config applies Channel changes, access identities and sessionAutoTitle immediately. Startup settings, such as the HTTP port, Node providers, semantic search and the inbound MCP listener, need a restart. Model changes in Setup → Models never need one.

With Docker, changing the port also means updating the port mapping (and, with Compose, the healthcheck). See Docker.

Back up the data directory, then run foxwarmctl update in the program directory, or pull the new container image. See Update a release installation.

The whole data directory (foxwarm-data by default), taken while Foxwarm is stopped or with a SQLite-consistent backup. See What to back up.

On the issue tracker. Include the Foxwarm commit, platform, install method, the error and steps to reproduce. Remove tokens, API keys, private messages and personal paths from logs first.