AI evidence infrastructure · Open protocol
We build AI that can
prove its work.
Phatfella is a small UK team building open evidence infrastructure for AI. Every product we ship runs on AIEP — the open protocol where every AI claim is backed by retrieved, hash-committed, independently replayable evidence.
The protocol
AIEP — the open standard for verifiable AI records
Most AI systems produce answers you cannot verify. AIEP changes that structurally.
Every AIEP response carries a hash-committed evidence trail — sources retrieved, SHA-256 hashed and locked at answer time, then published at /.well-known/aiep/ on any domain. No retroactive changes.
The standard is open. Any system can publish conformant records. Any AI retrieval agent can discover, parse and trust them. Schemas, GENOME constitutional SDK, reference validator and mirror toolkit — all open source. Protocol IP patents pending.
Hash-committed
SHA-256 at publish time
Evidence-bound
Every claim links to sources
Machine-readable
/.well-known/aiep/
Replayable
Any session, from first principles
How it works
Six properties of every AIEP record
Evidence-bound
Every claim links to the artefacts that support it — as a machine-readable, hash-verified field, not a footnote.
Hash-verified
Every artefact carries a SHA-256 commitment at the moment of publication. Changes after the fact are detectable.
Machine-readable
Published at /.well-known/aiep/ on any domain. AI retrieval systems can find, parse, and trust the content directly.
Dissent-preserving
Outlier claims are archived with their evidence intact — ready for recall when knowledge shifts.
Replayable
Every reasoning chain is committed to ledger. Any session, any inference, replayable from first principles.
Fail-closed
The GENOME kernel suppresses execution if any invariant is unmet. It does not attempt to continue with degraded integrity.
What we have built
Applications built on AIEP
Six products sharing the same verifiable evidence substrate.
Piea
Evidence-governed AI
Professional-grade AI with a verifiable source rail. Every answer is backed by retrieved, hash-committed evidence. Four-model consensus on every inference.
AIEP Proof
Evidence management
Capture every quote, agreement and promise before it's disputed. Browser extension, Gmail, Outlook and mobile share target — SHA-256 tamper detection on every file.
Forecast
Construction intelligence
Contract admin and project control built on AIEP. Every instruction, variation, RFI and cost event — evidenced, timestamped and audit-ready.
AIEP Miner
Governed ASIC fleet control
Open-source ASIC fleet automation with a full AIEP evidence ledger. Every decision recorded, every suppression proven. No black boxes — Apache 2.0.
Academy OS
Sports academy management
Members, gradings, billing and class scheduling. Evidence-backed attendance and belt progression records on the AIEP substrate.
Qardl
Verifiable professional identity
Credential fraud is structurally harder when every claim is cryptographically anchored. For regulated industries, built on AIEP.
Mirror
Make your domain AI-discoverable.
AI systems retrieve content from the web. Without a structured machine interface, your domain is invisible to AI retrieval — or at best scraped and guessed at. An AIEP Mirror is the fix: five JSON files at a known path, published on any web server.
Without a Mirror
- ✕ AI retrieval systems scrape or skip your domain
- ✕ Content cited by name but not verified
- ✕ No standard way to assert what you publish
- ✕ Cannot participate in federated evidence networks
With a Mirror
- ✓ AI systems discover and retrieve your artefacts directly
- ✓ Every artefact is hash-bound — changes are detectable
- ✓ metadata.json declares your identity machine-readably
- ✓ Other Mirrors can federate with yours
Who should publish a Mirror
Adoption tiers
- index.json
- metadata.json
Publisher identity and a basic discovery root. Enough for AI systems to find and recognise your domain.
- index.json
- metadata.json
- protocol.json
- crawler-brief.json
Add capability declarations and a crawler brief. AI systems can understand what you publish and how to retrieve it.
- All five files
- + artefact entries
- + federation config
Full artefact index with hash-bound entries. Participate in federated evidence networks. Verifiable by any AIEP-compliant system.
phatfella.com publishes a Mirror
Discovery artefacts are live at /.well-known/aiep/. AI retrieval systems can verify our identity without scraping.
Open source
The protocol infrastructure is public. Build on it.
Open-source ASIC fleet automation with full AIEP evidence ledger
Protocol hub, canonical schemas and machine discovery endpoints
Make any domain AI-discoverable in five JSON files — no backend
Open-source validator for AIEP-compliant domains and responses
Developer SDK for building AIEP-native applications
Evidence vector store and semantic retrieval reference impl
Build on AIEP.
The protocol is open. The infrastructure is public. Start with the GENOME SDK or fork any repo.