Context.
The whole market in one call. No auth. CC-BY-4.0.
3,000+ prediction markets across Kalshi and Polymarket, condensed by an LLM into eighteen topic groups, cross- category highlights, traditional-market snapshot, live thesis edges, and evaluation signals. A few hundred tokens — drop the JSON straight into your agent prompt.
Catalan Atlas · Cresques 1375 — geography, politics, astronomy, and trade routes pressed into one parchment. The original world-context bundle.
What ships in one call
Seven sections. ~few-hundred tokens total. No auth.
The bundle is shaped for an agent's next decision, not for human browsing. Categories rank by 24h volume, highlights surface what cannot be missed, edges and signals come from the active thesis loop.
categories18 topic groups (macro, crypto, policy, sports), each with topMovers + mostLiquid lists.
highlightsLLM-curated cross-category callouts with related tickers and suggested commands.
traditionalReal-time SPY, VIX, gold, oil, BTC snapshot — context for prediction-market repricing.
edgesLive thesis-vs-market edges from active public theses, depth-scored, top 8.
signalsRecent thesis evaluations: confidence delta, summary, citation per signal.
metaTotal markets scanned · K/PM split · scan timestamp · LLM token budget.
licensingCC-BY-4.0; attribute simplefunctions.dev. Drop the JSON straight into agent prompts.
Live thesis edges
Recent thesis evaluations
Why a curated context beats raw markets
Bloomberg dashboard
Best curated market screen money buys
No prediction-market integration; no agent-readable shape; no thesis context
Raw markets dump
Total coverage
Too large for agent context; no curation; no actionable framing
Custom scraper
Full control over the schema
Months of work; no LLM curation; no calibration; no settlement awareness
SimpleFunctions context
One call · LLM-curated · agent-shaped
Designed for the next tick decision, not for browsing
Six patterns that consume context
Same JSON; different downstream. The shape that repeats: an agent or a human reading the bundle and deciding what to do next.
Agent prompts
Embed sf.context JSON in the system prompt; agent reasons about today's tape.
Morning briefing
CLI sf context piped to your IC pack — categories + highlights + signals.
Slack / TG bots
Highlights → Slack message every heartbeat; only the changed callouts surface.
Research pipelines
Backtest replay context as features alongside per-tick orderbook history.
Public dashboards
CC-BY-4.0 — drop into your own dashboard; attribute simplefunctions.dev.
LLM training corpora
Daily HF dump of context bundles — clean signal for training prediction-market agents.
Context endpoints
Same JSON across CLI, REST/Data API, MCP adapter, and historical snapshots. Drop into any agent or pipeline.
FAQ
What does the global context endpoint return?
A curated bundle: categories (with top movers and most-liquid contracts per category), cross-category highlights (with actionable tickers and suggested commands), traditional market snapshot (SPY, VIX, gold, oil, BTC), live thesis edges, recent evaluation signals. ~3,000 markets scanned, condensed to a few hundred tokens designed to drop into an agent prompt.
How is this different from /api/public/markets?
markets is the raw catalogue. context is the curated, decision-shaped subset: only what an agent or PM should react to right now. Highlights have plain-English descriptions and ticker links; categories rank by 24h volume; edges and signals come from active public theses.
How often is it refreshed?
Categories and highlights are LLM-curated twice daily. Edges and signals refresh on every heartbeat tick (15 min). Traditional market snapshot is real-time. The endpoint cache is ~30s; pull it as fast as you like.
Do I need an API key?
No. /api/public/context is CC-BY-4.0 licensed and available on the free tier with rate limits. Authenticated thesis-specific context (per-thesis tree state) is metered and key-gated.
How do I drop this into an agent?
Three lines: agent reads sf.context (or GET /api/public/context), embeds the JSON in its prompt, takes action via sf.intent.create. The bundle fits in ~few hundred tokens; agents can pull on every loop without context-window pressure.
What are highlights actually?
LLM-curated cross-category callouts: an event the system thinks the agent should not miss (a Fed pivot, a CPI surprise, a Polymarket sports liquidity hot spot). Each highlight ships with related tickers and a suggested follow-up command.
Categories — by what taxonomy?
Eighteen topic groups: macro (oil, Fed, CPI, recession, treasury yields, geopolitics, tariffs, elections, govt shutdowns, central banks), crypto (BTC/ETH/SOL ladders + ETF flow + halving), policy (legislation, executive actions, regulatory), and sports (EPL, NBA, UCL, CS2, IPL, NFL). Each category gets its own movers + liquid list.
How is global context different from thesis context?
Global context shows the market as a whole — categories, movers, highlights. Thesis context layers your model on top: causal tree with thesis-implied prices, edges where your view disagrees with the market, kill-condition status, track record. Global is no-auth; thesis is BYOK.
Cost?
Public global context endpoint is free, CC-BY-4.0. Per-thesis context is metered. Webhook delivery free. WebSocket /v1/ws free for read.
Backtest / historical?
GET /api/public/context?at=2026-04-01T12:00 returns the historical snapshot at that timestamp. Daily Hugging Face dump under SimpleFunctions/context (CC-BY-4.0) ships the full archive.
Related surfaces
Heartbeat engine
How context refreshes — 15-min ticks plus out-of-band on event.
World state
Single agent-sized snapshot — even tighter than full context.
Edge discovery
How thesis-vs-market edges in the context bundle are computed.
OpenClaw skill
Add the MCP skill, agent gets context.simplefunctions natively.
Macro context
Macro-only slice with thesis, edges, kill chain.
Policy context
Policy-only slice with legislation, executive actions, regulatory.
Prediction market API
CLI, REST/Data API, real-time WebSocket streams, MCP adapter.