How to Use SimpleFunctions

Six scenarios, six commands. Pick the one that matches your problem.

1. Your AI agent doesn't know what happened today

curl -s simplefunctions.dev/api/agent/world

800 tokens. The state of 34,000+ live Kalshi and Polymarket contracts, liquidity-weighted, refreshed every 15 minutes, free.

What it looks like right now (April 24, 07:30 UTC):

Regime: Broad up-skew — markets resolving toward higher probability
SF Index: Disagree 40 | GeoRisk 27 | Breadth +0.37 | Activity 100

[divergence] Russia entering Malokaterynivka by May 31 overshooting Orikhiv — 30¢ gap
[residual]   3.5σ volume at 6¢ on "CPI rise more than 0.6% in May"
[divergence] Iran military action overshooting uranium deal — 16¢ gap
[consensus_break] WTI front-month settle deeply split across 6 contracts (σ=43¢)

Paste into a system prompt and Claude Haiku 4.5's accuracy on world questions jumps from 2.3% to 70.5%. The benchmark is public.

For deltas only: sf world --delta — 30-50 tokens of "what changed since last check."

For regional drilldowns: sf world iran or /api/agent/world/iran. Eight default regions — iran (278 markets), oil (426), fed-rate (1,273), recession (688), election-2026 (2,830), bitcoin (1,182), ukraine (324), china (385) — or any topic slug.

2. Your user asks a specific question

"Will the Fed cut rates in June?"

curl -s simplefunctions.dev/answer/fed-rate-cut-june-2026

22% — Will the Fed cut rates in June 2026?
Polymarket 18% · Kalshi 27% · 20 contracts · $154K volume · high confidence
Cross-venue gap: 9pp (Kalshi higher)

Contracts:
- How many Fed rate cuts in 2026?: 0 (0 bps) — 41¢ Polymarket $72K
- How many Fed rate cuts in 2026?: 4 (100 bps) — 5¢ Polymarket $25K
- How many Fed rate cuts in 2026?: 1 (25 bps) — 28¢ Polymarket $15K
  ...

3,200+ topics, liquidity-weighted across every contract that touches each one. Same URL serves markdown to agents and a Wikipedia-styled page to humans. Cite as: "22% per prediction markets (SimpleFunctions, April 2026)."

3. You want to filter contracts

34,907 live contracts. You want the ones that matter.

sf screen --iy-min 80

Every contract with implied yield above 80% annualized. Right now: 19,695 matches post-filter, sorted by yield.

sf screen --cri-min 10

Cliff-risk filter — price has moved sharply relative to its own volatility. Somebody knows something, or somebody panicked.

sf cross-venue

Kalshi-Polymarket pairs where the same binary outcome trades at different prices. Pipeline matches them via FTS plus LLM semantic checks. The gap is the signal.

Every command accepts --json for pipes.

4. You have a view

Not just watching. You think oil stays above $100.

sf create "Oil stays above $100 for six months"

Your sentence becomes a causal tree: the claim at the top, sub-claims below (supply tightness, OPEC discipline, geopolitical risk), measurable leaves at the bottom. Every leaf links to live contracts.

A heartbeat runs every 15 minutes — rescans news, refreshes prices, re-evaluates the tree, writes confidence deltas. When a sub-claim breaks, you see which leaf dragged it. Not "oil dropped," but "OPEC-cuts-persist-through-Q3 dropped from 82¢ to 61¢."

sf edges

Thesis-implied price minus market price, minus half-spread, across every leaf. In cents, logit space, or annualized. That gap is your edge.

5. You want the agent to trade

sf portfolio enable
sf portfolio view add "Recession risk is underpriced" "Curve and payrolls diverge..." --conviction 4
sf portfolio trigger

The agent reads your views, scans the market, finds contracts that express them, and runs every candidate through seven mechanical risk gates — position limit, concentration, correlation, liquidity depth, cliff risk, overround sanity, drawdown budget. No LLM can override the gates.

Dry-run by default. sf portfolio last shows what the agent would have done. When you're ready:

sf portfolio config execution_mode live

We run live theses of our own. Every one is public at simplefunctions.dev/theses — the tree, the confidence trajectory, the positions, the parts that were wrong. The point is that the wrong parts are visible.

6. You're building agents

claude mcp add simplefunctions https://simplefunctions.dev/api/mcp/mcp

56 tools. Same verbs as the CLI: scan, screen, context, watch, cross-venue, yield-curve, calibration, regime, calendar, contagion, inspect. Claude Code, Cursor, Cline.

Or REST: 60 endpoints at /openapi.json.

Or Python: pip install simplefunctions-ai.

Or shell: everything above is a curl.

Pricing

Free during beta. Pay-by-token after 15M tokens consumed. Developer-infrastructure pricing, not SaaS.

Start here