Run it on a server
The local install suits most players. For an agent that plays even when your computer is off, run it on a small
Linux server (Ubuntu with systemd, on an x86-64 processor). The repository’s deploy/ folder has everything; the full guide is
docs/SERVER.md in the repository.
github.com/foxsyai/foxleague-agent public at the launch
In short:
- Copy the repository to the server and run, as root,
deploy/bootstrap.sh: it creates a dedicated user, a private Node.js, the folders and a sandboxed systemd service. - From your computer, ship the code:
DEPLOY_HOST=root@your-server ./deploy/deploy.sh --no-restart. - On the server:
agentctl.sh setup(the wallet),agentctl.sh configure(keys; a 6-digit code links your Telegram),agentctl.sh set-web-password(the dashboard login),agentctl.sh check-keys. - Start it:
systemctl enable --now foxleague-agent. - Optional: the dashboard on your own domain behind nginx with TLS, from the template
deploy/nginx-foxleague-agent.conf. Set up from an older template, raise itsproxy_read_timeoutto1080s: Talk to the manager can take up to 17 minutes at the highest effort.
A server set up this way does not update itself. The dashboard and Telegram still tell you when a new version is
out; ship it from your computer with ./deploy/deploy.sh, as in step 2 (Update).
On a server the recovery words are read once in your own SSH session and written on paper; the agent refuses to
print them there. PEM files are written on your own computer, never on the server: with the same recovery words
in your local copy’s .env, ./deploy/sync-pems.sh writes a pem/agentN.pem for every agent the server runs
(it checks the addresses first). The dashboard reminds you each time you add an agent.
The full guide also says how to let one or more other AI models plan in the shadow, to compare them with the one
that plays before you switch (What it costs). It shows how
to read the comparison with agentctl.sh shadow-report too, how to set an agent’s own model and effort from the
command line (agentctl.sh model), how to let one agent’s model plan in the shadow at other efforts and how to let
the model that plays plan once more at its own effort, as a baseline for the others.