anjeer 0.1.0

CLI for the Anjeer trading game — find, create and join API-mode lobbies

CLI & Scripted Players

A Python CLI for managing lobbies and launching scripted players in API-mode games.

Installation

pip install -e cli/          # RECOMMENDED: dev install from repo root
# or
uv tool install anjeer       # production install

Run anjeer setup from your bot's project directory. It prompts for your API key and preferred language, saves config to anjeer.json in the current directory, and copies a starter script template.

Commands

anjeer setup                              # first-time setup in cwd
anjeer find                               # list open API-mode lobbies
anjeer create                             # interactive lobby creation
anjeer create --game-mode advanced \
              --spawn-bots \
              --bot-difficulty hard       # non-interactive
anjeer join <code>                        # join a lobby and launch your script on start
anjeer start <code>                       # (creator) start the game and launch your script
anjeer close <code>                       # leave and delete if empty
anjeer preset list/save/delete            # manage saved lobby configurations

anjeer start and anjeer join set environment variables (ANJEER_API_KEY, ANJEER_SERVER_WS_URL, ANJEER_LOBBY_CODE, ANJEER_GAME_MODE) then exec your configured script. Your script connects to the same WebSocket the browser uses and receives the full wire protocol.

Scripted Players

Scripts receive everything: MBP-N depth snapshots, MBO order lifecycle events, eval module outputs, delta tables, and round lifecycle messages. This is enough data surface to build serious strategies.

The spectator view shows a live script log panel — your script can send script_log messages over the WebSocket and they appear in real time for anyone watching.

A few directions people might take this:

• Bayesian agent — replicate or extend the hard bot's hypergeometric posterior in Python with full access to your own hand
• Time-series model — buffer depth snapshots and trade events into a feature matrix; act on confidence crossings
• Reinforcement learning — treat each round as an episode; reward = end-of-round balance delta

The interface is stable enough for real strategies. Helper templates covering connection setup, book reconstruction, and hand tracking are planned.