Product

Your AI can already think about the future. Credence gives AIs the data to do it right.

Our probability API grounds your AI in real-world evidence and probabilities — backed by financial markets, not hallucinations.

One query. One probability object.

Ask “How likely is event X?” and you get three numbers that reconstruct the full posterior:

  • p_market — the raw prediction-market midpoint (what the crowd prices).
  • p_calibrated — our refined probability for the same event.
  • n_effective — evidence strength: the Beta posterior concentration, i.e., how much to trust p_calibrated.

From the last two you get the entire Beta posterior — alpha = p_calibrated × n_effective, beta = (1 − p_calibrated) × n_effective — not just a point.

A syntax-highlighted /v1/probabilities/query JSON response with snapshot lineage and the probability object.
A single probability object, with audit-grade lineage.
Two posteriors at the same probability — one with high evidence strength, one with low.
Same probability — the evidence strength is the difference.

Same probability, different decisions.

A price approximates the odds. It doesn’t tell you the evidence strength behind them — n_effective does.

Two events can sit at the same probability, one with high evidence strength and one with low evidence strength. Your agent should lean in on the high-evidence-strength posterior and hold back on the low one. That’s the signal a price can’t carry.

See it flip real decisions

Built for agents. REST when you need it.

Credence is MCP-native — your agent calls it as a tool inside its own loop, no glue code. Or hit REST directly. Same probability object either way.

Integrates easily via MCP with Anthropic, OpenAI, Gemini, Copilot, and more.

From messy markets to a clean distribution.

We mirror every open market continuously, preserve the raw tape, and train models over it to produce p_calibrated and n_effective (evidence strength).

The market aggregates the information; we read the calibrated probability and evidence strength back out — every tick.

Read the methodology

Every market, every tick.

Credence covers all Polymarket markets tick-by-tick. Kalshi is next. No cherry-picked sample: if it’s live, it’s covered.

< 0.1 sec
API latency (excluding network)
70,000+
live assets tracked
< 1 minute
freshness of served data
all markets, all ticks
feed our live state

Every number traces back.

Each response carries audit-grade lineage — snapshot ID, model version, release, and inputs. Pin a probability to an immutable snapshot and reproduce it later by digest.

Bloomberg-grade provenance on market-implied probability objects.

Let your agent see the future.

We’re working with a small number of design partners ahead of public launch.