The runtime admissibility layer for AI in high-stakes decisions.

Website · Docs · Live service · Demos · Trust portal · Press

Every governed action ships a Decision Receipt — a cryptographically signed (Ed25519), replayable, independently verifiable record of why an answer was allowed, revised, blocked, or escalated.

Authorized is not the same as admissible. An autonomous system can be permitted to act and still owe an account of how it decided. Summit produces that account at runtime, as the decision happens, rather than reconstructing it under pressure months later.

A receipt binds four things into one tamper-evident artifact, chained to the receipt before it:

• Evidence — typed, provenanced sources: CI results, code review, test output, human approval, model traces.
• Policy — the rules evaluated and the verdict they produced.
• Replay — whether the same inputs deterministically reproduce the same outcome.
• Attestation — the Ed25519 signature and hash chain, verifiable offline against a published public key.

# Service health — no auth
curl -s https://decrec.summitcognitive.ai/health

# The public key receipts verify against — no auth
curl -s https://decrec.summitcognitive.ai/v1/keys/server

# Free-tier API key (100 requests/hour)
curl -s https://decrec.summitcognitive.ai/v1/signup \
  -H 'Content-Type: application/json' \
  -d '{"email":"you@example.com"}' | jq -r .api_key

• Quickstart — docs.summitcognitive.ai
• API reference — docs.summitcognitive.ai/api/reference
• Examples & recipes — docs.summitcognitive.ai/guides/examples

Teams whose automated decisions get reviewed by someone who was not in the room: regulated industries, public sector and defense, critical infrastructure, and any operator whose agents take consequential action without a human in the loop on every step.

The EU AI Act's record-keeping and traceability duties — Articles 12, 19 and 26(6) — are architectural obligations, not retention-shelf items. Instrumenting an evidence layer while a system is being built is materially cheaper than retrofitting one underneath a running agent network.

A receipt that verifies is proof about the record — that the inputs, the policy and the outcome are what they claim to be, unaltered and in order.

It is not a warranty that the model was correct. Correctness remains a human judgement. What changes is that the judgement becomes reviewable instead of unreviewable.

Enterprise pilots and regulated deployments, including the 30-day Decision Admissibility Pilot: brian@summitcognitive.ai

_{Summit Cognitive, Inc. — Delaware C-corporation, founded April 2026. Alma, Colorado. Founded by Brian Long.}