Arena Open Ladder

What you submit

A Python agent: class MyAgent(RemoteAgent) with on_tick(state), called every tick. Return a list of orders to place or cancel. You connect over WebSocket and trade in real time against the background market and other players — there is no file to upload.

pip install convexpi-arena

from convexpi.arena import RemoteAgent, MarketState

class MyAgent(RemoteAgent):
    def on_tick(self, state: MarketState):
        if state.mid:
            return [self.limit("buy", round(state.mid) - 5, 10)]
        return []

MyAgent("your-handle", server="wss://…").run()

state carries the mid, the order book, and your position/PnL. Return limit/market orders or cancels each tick.

How you’re scored

Ranked by Realized profit and loss (PnL).

• Your agent posts and takes orders on a real order-by-order (L3) book; queue position and fills are genuine.
• You trade against background agents and other players; maker/taker fees apply.
• You are ranked by realised PnL, updated live every few ticks.

The data

A live limit-order book — for the realistic exchange, replayed order-by-order from real recorded crypto market data, so the queue you join and the trades that fill you are real.

Getting the data you fit on: The starter notebook connects a baseline RemoteAgent to the live book so you can watch the state and build from a working quote.

Approaches to try

• Start from the baseline market maker: quote both sides around the mid and earn the spread.
• Skew or widen your quotes as inventory builds, to limit adverse selection.
• React to the book — pull quotes when the market moves against you (win the cancel race).

Just a foothold — the missions and starter notebook go deeper.

Timeline & rules

Get started

The Arena server is being set up — live play opens shortly. Watch the rankings in the meantime.