The denoised, entity-resolved market-fact layer for AI agents.
Your agent reasons over clean, typed facts - one canonical ID per company and coin across equities and crypto, dated canonical events and a knowledge graph - instead of parsing a raw-headline firehose. It is the ETL you would otherwise build yourself.
Not another raw-data firehose - the layer you read first
Everyone sells raw prices, sentiment scores, or human-written research. We sell the clean substrate an agent consults before any of them: raw feeds turned into structured, entity-resolved, dated facts that are cheap to reason over.
Typed facts, not headlines
Every item carries an event type, a resolved subject ticker, a USD magnitude where present, a direction hint and a source URL. Generic headlines are dropped at the source.
One canonical identity
NVDA, Nvidia, @nvidia and $NVDA resolve to a single ticker - across both equities and crypto. Deduplicate the whole feed for free. Precision over recall: null before a wrong merge.
Dated canonical events
Structurally important events graded with a p_canonical score - the early-warning layer that flags an inflection before consensus, dated and traceable. Provenance of timing, not a return claim.
We sell convenience and provenance - not alpha
This is informational market intelligence, not investment advice and not a claimed price edge. Every forward call is logged before its horizon and graded openly - hit-rate, Brier, calibration, control group and the global false-discovery rate all disclosed. You can check the numbers yourself.
Pricing
Three cumulative tiers - a higher tier includes everything below it. Every request is metered. Start free and see the full scope before you pay.
See the scope before you commit.
- Full scope in one call - datapoints, sources, tracked actors, dated forward-calls (
/api/scope) - Self-describing API - machine-readable catalog (
/api/docs, no key)
The active trader who wants verifiable, dated intelligence.
- Honest forward track record - hit-rate, Brier, calibration (
/api/track) - Canonical-event early warning with
p_canonical(/api/canonik) - Typed, entity-resolved facts + one canonical ID per name (
/api/facts,/api/resolve) - Dated evidence dossier + money-flow rails (
/api/dossier,/api/govawards)
The agent / quant who wants the whole structured feed.
- Everything in Signal, plus:
- Full knowledge graph + per-node dossiers (
/api/graphnet,/api/graphnode) - World-compass scenarios - winners, losers, safe havens (
/api/weltkompass) - Point-in-time fundamentals from SEC XBRL (
/api/fundamentals) - Regime, risk, alliances, industry map and the learned knowledge layer
Built for machines first
Fetch the OpenAPI 3.1 spec and generate a client or an MCP server from it. Send your key in the X-API-Key header. Errors are honest - 401/403/429 with Retry-After.
Entry points
OpenAPI 3.1 spec · JSON self-description · agent skill file · llms.txt
Recommended flow
GET /api/scope → GET /api/facts?subject=NVDA → GET /api/graphnet → GET /api/graphnode?id=...
Frequently asked
What is the Market Signal API?
It is a denoised, entity-resolved market-fact layer for AI agents. It turns the raw stream of financial news and tracked-actor posts into typed, deduplicated facts - one canonical ID per company and coin across both equities and crypto - plus dated canonical events, a knowledge graph and a crypto hype-velocity detector. It is the ETL an agent would otherwise build itself.
Who is it for?
AI trading, research and optimization agents, and the developers and quants who provision them. Your agent reasons over clean, structured facts instead of parsing a raw-headline firehose.
How is it different from a raw market-data API like Polygon or CoinGecko?
We do not compete with them - we are the layer you read before them. Raw-data APIs give you prices and headlines; we give you denoised, entity-resolved facts and events on top, so the raw firehose becomes cheap to reason over. Think of us as the substrate an agent consults first.
Does it give trading signals or claim market-beating alpha?
No. This is informational market intelligence - convenience and provenance, not investment advice and not a claimed price edge. Every forward call is logged before its horizon and graded honestly (hit-rate, Brier, calibration) at /track. We sell the clean substrate, not a promise.
How much noise is removed?
About 91% of generic headlines are dropped at ingestion - a recent run distilled 28,179 raw items to 2,402 typed facts with zero false positives in review. Only facts that carry a type, a resolved subject and (where present) a magnitude survive.
What is entity resolution and why does it matter for an agent?
One canonical identity per company or coin: NVDA, Nvidia, @nvidia and $NVDA all resolve to a single ticker. An agent can deduplicate entities across the whole feed instead of treating four spellings as four things. We favor precision over recall - it returns null rather than a wrong merge.