chipzen-bot

The Python adapter for the Chipzen AI poker competition platform. Wraps the WebSocket protocol so your bot only has to implement decide(state) -> action, and ships a validate CLI that confirms your bot will be accepted by the upload pipeline.

Install

pip install chipzen-bot

Python 3.10+ is supported. The runtime dependency is a single package (websockets); your bot can pull in whatever else it needs (numpy, torch, etc.) on top.

Minimal bot

import asyncio
import os
from chipzen import Bot, Action, GameState, run_bot


class MyBot(Bot):
    def decide(self, state: GameState) -> Action:
        if "check" in state.valid_actions:
            return Action.check()
        return Action.fold()


if __name__ == "__main__":
    # An uploaded bot gets its match URL from the platform via $CHIPZEN_WS_URL.
    asyncio.run(
        run_bot(os.environ["CHIPZEN_WS_URL"], MyBot(), token=os.environ.get("CHIPZEN_TOKEN"))
    )

To run a bot remotely from your own machine instead of uploading it, use run_external_bot(...) / chipzen run-external — see Two ways to run a bot.

The SDK handles the Layer-1 transport handshake, Layer-2 game-state parsing, ping/pong, request-id echoing, action_rejected retries, and reconnect. Subclass Bot, override decide(), return an Action. That's the entire surface for a working bot.

Lifecycle hooks (on_match_start, on_round_start, on_phase_change, on_turn_result, on_round_result, on_match_end, on_decision_latency) are optional — override them if you need to maintain per-match or per-hand state between turns or log your decision timings.

Two ways to run a bot

The same Bot class works on both paths:

• Upload (containerized). Package your bot as an image and submit it; the platform's executor runs it. This is the run_bot(...) / chipzen-sdk validate + Docker path above — best for ranked competition and tournaments.

• External-API (remote play). Run your bot on your own machine and let the platform match and dispatch it over the public token-authed API — no upload, fast iteration:

  import asyncio
  from chipzen import Bot, run_external_bot
  
  asyncio.run(run_external_bot(MyBot(), bot_id="<bot-uuid>", env="prod", token="cz_extbot_..."))
  

  It holds one lobby connection and plays every match dispatched to your bot — a single challenge, or each round of a tournament. Put the token in a chipzen.toml ([external_api] token = "cz_extbot_...", optional bot_id / url) and the CLI is a one-liner:

  chipzen run-external my_bot.py   # targets prod (chipzen.ai) by default
  

  Tunables: connect_to_chipzen() (env→URL), RetryPolicy (reconnect/backoff), safe_mode=False (crash on a decide() bug instead of folding — for dev/eval). See docs/external-api/FIRST-30-MINUTES.md.

CLI

The chipzen-sdk CLI (aliased as chipzen) is installed alongside the package:

Run chipzen-sdk <command> --help for the full option list per command.

What the SDK is for (and what it isn't)

The SDK does three things and nothing else:

1. Protocol adapter — your bot doesn't hand-roll WebSockets.
2. chipzen-sdk validate — pre-upload conformance check.
3. IP-protected Dockerfile recipe — the Cython multi-stage build that ships in starters/python/Dockerfile (cythonize -i bot.py && rm bot.py) produces an image containing only compiled .so files, not your .py source.

It does not include a local match simulator, hand evaluator, or opponent pool. Upload review only checks that your bot speaks the protocol (a short scripted mini-match); you learn how strong it is from the real matches it plays on the platform (challenges, tournaments and seasons, with ratings and replays). If you want fast local iteration, write your own profiling harness that calls your Bot.decide() directly with recorded GameState objects.

Reference

Full developer documentation lives in the chipzen-sdk repo:

• DEV-MANUAL — SDK reference, lifecycle hooks, performance budgets, container contract, troubleshooting.
• QUICKSTART — write a bot, validate it, package it (~10 minutes).
• Protocol spec — Layer 1 (Transport) + Layer 2 (Poker game state). Authoritative.
• Bot runtime security model — what the platform enforces on uploaded bots (sandbox, network egress, resource limits).

Per-package quickstart: QUICKSTART.md.

License

Apache-2.0. See the LICENSE file in the repo.