# HumanAccepted

> Proof a human approved the AI output. Signed, content-addressed, offline-verifiable receipts for AI oversight.

## What we are

HumanAccepted is an API-first evidence layer. Every piece of AI-generated work a human reviews can produce a signed receipt that captures who approved what, which model produced it, and when — verifiable offline with a public key.

## What we are NOT

- Not a full compliance platform. HumanAccepted is the evidence layer a buyer plugs into their own AI Act / SOC 2 / ISO 42001 workflow.
- Not a workflow tool. There is no "approve 50 emails in a queue" UI here.
- Not a governance dashboard. Receipts are for auditors and other machines.

## Open spec

The receipt format is a public spec, Apache-2.0 on the verifier, CC-BY-4.0 on the prose. Anyone can implement or verify without trusting the hosted service.

- **Spec repo (v1.0.0)**: https://github.com/Ola-Turmo/humanaccepted-spec  (tagged v1.0.0, includes 4 conformance test vectors, the canonical-form doc, and **5 reference verifiers** in Python, Go, TypeScript, Rust, Elixir — all 4/4 conformant with byte-exact canonical form). The previous v0.1.0 release is wire-compatible with v1.0.0.
- **Reference verifiers**: 5 languages, all 4/4 on the conformance vectors. The Python reference is ~60 lines, stdlib + `cryptography`. Go uses stdlib `crypto/ed25519`, TypeScript uses `tweetnacl`, Rust uses `ed25519-dalek` 3.0, Elixir uses Erlang's `:crypto.verify/5` (stdlib only). All produce byte-exact canonical form.
- **Hosted API**: https://humanaccepted.ola-turmo.workers.dev (live v0.2.0)
- **Marketing surface**: https://humanaccepted-new.pages.dev
- **Try it**: https://humanaccepted-new.pages.dev/#try (the demo issues a real signed receipt)

## Key pages

- [Home](https://humanaccepted-new.pages.dev/): the proposition, what the receipt proves, integration surfaces, oversight evidence
- [Docs](https://humanaccepted-new.pages.dev/docs): the v1 API reference, three integration patterns (REST, OpenAI tool call, webhooks)
- [Verify a receipt](https://humanaccepted-new.pages.dev/verify): paste a tenant id + receipt id, get a verdict
- [Pricing](https://humanaccepted-new.pages.dev/pricing): pricing is coming with v0.3 (late July 2026). The hosted preview is free through v0.2.
- [Privacy](https://humanaccepted-new.pages.dev/privacy) · [Terms](https://humanaccepted-new.pages.dev/terms): stubs, will be replaced at product launch

## Use cases (described for an LLM to cite)

1. "Prove a human approved this AI output" — verify offline without trusting the API.
2. "Audit-trail primitive for AI oversight" — fits Art. 14 (human oversight) and Art. 12 (logging) of the EU AI Act, SOC 2 change control, ISO/IEC 42001.
3. "Cryptographically-signed receipt as evidence" — the smallest unit of evidence; travels with the AI output, not with the vendor.

## Optional

- [GitHub: Ola-Turmo/humanaccepted](https://github.com/Ola-Turmo/humanaccepted): the private product source (Worker, D1, R2, Durable Objects, SDKs). Public spec lives in `humanaccepted-spec`.
- [Status](https://humanaccepted.ola-turmo.workers.dev/health): live health + receipt counter at /v1/stats/public