Predict.fun API Guide: Connect a Trading Bot in 5 Steps

Auth: API key issued in the account settings, sent as a header.
Read: REST for markets and ladders; a book stream for live updates.
Trade: Limit and taker orders per share, tick 0.01.
Taker fee: 0.018 x min(p, 1-p) per share, charged in outcome shares. Makers pay nothing.
Settlement: A point read of the oracle price at each boundary. No averaging. Ties pay 50/50.
Geography: Blocks the US.
Docs: docs.predict.fun

Working with an AI assistant?

This brief tells an assistant how Predict.fun authenticates, what the fee formula is and where the traps are. Copy it, paste it, then ask your question.

Show the prompt

Help me connect a trading bot to Predict.fun's API. Context: the short crypto up/down markets settle on a point read of the Chainlink price stream at each window boundary (no averaging), windows chain exactly so one window's end price is the next window's start price, and ties pay 50/50. Show me: authenticating with an API key, listing the current 5-minute and 15-minute markets, reading the order book ladder, placing a taker buy, and reading my fills. Gotchas: the taker fee is 0.018 x min(p, 1-p) per share and is charged in outcome shares, makers pay nothing, the book is mostly one-sided in the last minute, and a 25 percent maker rebate plus points program has no announced cash value. Point me to the official documentation for exact endpoint paths. Source: thebotledger.com/api/predict-fun

Five steps, simplified. Endpoint names below are placeholders for the shape of each call; take the exact paths from the reference, which changes more often than the larger exchanges.

Authenticate

import requests
BASE = "https://api.predict.fun"          # confirm in the docs
H = {"X-API-Key": API_KEY}

List the current windows

markets = requests.get(f"{BASE}/markets", params={"category": "crypto", "status": "open"}, headers=H).json()
m = next(x for x in markets if x["timeframe"] == "5m" and x["asset"] == "BTC")
start_price = m["startPrice"]              # the declared open, exact to the tick

Windows chain: the end price of one window is the start price of the next. If your own spot feed disagrees with the declared start price, trust the declared one.

Read the ladder
```
book = requests.get(f"{BASE}/markets/{m['id']}/orderbook", headers=H).json()
best_ask = book["asks"][0]                 # [price, size]
best_bid = book["bids"][0]
```
In the last minute almost every print lifts the offer or hits the bid; nothing trades inside the spread. Model your fills at the touch, not the mid.

Place a taker buy

order = {"marketId": m["id"], "outcome": "UP", "side": "BUY", "type": "LIMIT",
         "price": best_ask[0], "size": 30, "timeInForce": "FOK"}
r = requests.post(f"{BASE}/orders", json=order, headers=H).json()

The fee comes out of the shares you receive, so a 30-share buy at 50 cents delivers about 29.7 shares.

Read fills

fills = requests.get(f"{BASE}/fills", headers=H).json()

Settlement is the difference

Polymarket, Limitless and Kalshi all settle short crypto markets on an average over the last minute. Predict.fun settles on a single read. On flat windows the two rules disagree about one time in ten, so do not pool Predict.fun outcomes with the others, and do not carry a model trained on the others across without retraining.