Get started

Install the sensei command, scaffold a bot, and play your first match — all on your machine. You bring your AI agent; sensei handles the rest. The whole thing takes a few minutes.

No account needed to start. Steps 1–3 run entirely on your machine. You only sign in at step 4, when you're ready to put a bot on the ladder.

Install the sensei CLI

This one line downloads our install script (with curl) and runs it (| sh). It drops a single sensei binary into your PATH — nothing else to set up.

curl -fsSL https://get.agentsensei.dev | sh

Cautious about piping to a shell? Read the script first, or download the binary from Releases and put it on your PATH yourself. Have Rust? git clone the repo and cargo build --release -p arena-cli.

Verify it's ready:

sensei doctor

Create your first bot

Scaffolds a starter ant in bot/ AND drops a coaching guide (AGENTS.md + .agents/) next to it — so when you open the folder in your AI agent, it already knows how to coach you from zero to a ranked bot. No separate clone needed.

sensei init --lang python

Or rust, go, typescript, cpp. New to this? Pick Python.

Add --open to launch it in your editor right away (or --ide cursor / code). Then tell your agent "get me started" — it reads the guide and takes it from there.

Play & watch locally

Runs a real match against a built-in opponent on the embedded referee (the exact engine the server uses), then opens a replay you can scrub through. This is your iteration loop — no account needed yet.

sensei match --against baseline

sensei replay latest

Join the ladder

Sign in (here on the site, or from the terminal), then submit. We build your bot, run it in ranked fog-of-war matches, and rank you on the leaderboard — every game replayable.

sensei login

sensei submit

Prefer the browser? Sign in here and your CLI session links automatically.

No API key? Build fully local

Don't have (or want to pay for) a cloud agent? One command checks your machine, picks a local model that fits, and installs a local runtime + agent — no key, offline, private.

sensei setup --local

Heads up: local models scale with your hardware — great on a strong machine, weaker than cloud agents like Claude on a 16 GB laptop. On low-spec hardware sensei will say so and point you back to the cloud path. Add --dry-run to see the plan first.

Stuck? Your AI agent already knows this flow — open the cloned repo and say "get me started". It reads the bundled guide and walks you through each step.