What We're Building
A trading bot that:
- Maintains a macro thesis (e.g., "US recession by end of 2026")
- Continuously scans Kalshi for contracts where market price diverges from thesis-implied value
- Alerts you (or trades automatically) when edges appear
- Tracks P&L and adjusts based on new information
This isn't a latency-sensitive HFT bot. Prediction markets reward information advantage, not speed. The bot's job is to systematically find and act on mispriced contracts.
Prerequisites
npm install -g @spfunctions/cli
sf setup
You'll need:
- SF API key (free at simplefunctions.dev/dashboard)
- Kalshi API key with read+write permissions
- OpenRouter API key (for the agent's LLM — used for evaluation, not trading logic)
sf setup walks you through all of these with real-time validation.
Step 1: Create Your Thesis
sf create "US enters a recession by end of 2026 — tariffs drive supply chain costs, inflation stays sticky, Fed can't cut fast enough"
This takes ~30 seconds. The engine:
- Decomposes your thesis into a causal tree of testable sub-claims
- Scans Kalshi and Polymarket for contracts matching each node
- Calculates edges — where market price ≠ thesis-implied price
- Sets initial confidence
Step 2: Monitor with the Heartbeat
The heartbeat engine runs server-side every 15 minutes:
# Check what the heartbeat found
sf context <thesis-id>
# See top edges across all theses
sf edges
# Detailed evaluation history
sf feed --hours 24
You don't need to build a cron job. The server handles news scanning, price refreshing, and milestone detection automatically.
Step 3: The Trading Loop
For manual trading with bot-like efficiency:
# See what's actionable
sf edges --json | jq '.[] | select(.executableEdge > 10)'
# Check liquidity before entering
sf liquidity --topic oil
# Place an order
sf buy KXRECSSNBER-26 100 --price 32
For agent-assisted trading:
sf agent <thesis-id>
> analyze my top edges and suggest trades
> buy 100 contracts of the recession market at 32 cents
The agent has 21+ tools including place_order and cancel_order. Every trade shows a preview and requires confirmation.
Step 4: Scenario Analysis
Before committing capital, test scenarios:
# What if the war ends?
sf whatif <thesis-id> --set "n1=0.1" --set "n3=0.2"
# What if oil hits ?
sf whatif <thesis-id> --set "n3.3=0.95"
This is pure computation — zero LLM cost, instant results. It shows you how your edges change under different scenarios.
Step 5: Monitor from Anywhere
# Terminal dashboard (Bloomberg-style)
sf dashboard
# Telegram bot (runs in background)
sf telegram --token YOUR_BOT_TOKEN --daemon
# Track performance over time
sf performance
The Telegram bot pushes alerts when confidence changes >2% and responds to natural language queries.
Architecture
Your thesis
↓
Causal tree decomposition (LLM, one-time)
↓
Market scan: Kalshi + Polymarket
↓
Edge detection: thesis price vs market price
↓
Heartbeat (every 15 min):
- News scan (Tavily)
- Price refresh (Kalshi orderbook)
- Milestone check (economic calendar)
- LLM evaluation → confidence update
↓
You: read context → decide → trade
The bot isn't making trading decisions — it's maintaining an up-to-date view of where markets disagree with your thesis. You decide when to act.
Cost
- SF API: Free tier covers 15M tokens/month
- OpenRouter: ~/bin/zsh.05 per evaluation (Claude Sonnet)
- Tavily news: Smart scheduling keeps this under /thesis/month
- Kalshi: Standard exchange fees
Total running cost for one active thesis: roughly -30/month.
For the complete API reference, see simplefunctions.dev/docs.
