THE PLATFORM
Memory, built
to leave evidence.
A persistent evidence layer for AI systems. Capture the record, preserve its lineage and recover the context that was available.
CAPABILITIES
The primitives of a reviewable memory.
Evidence-first history
Record the source, context and configuration behind consequential AI outcomes.
Explore capabilityCryptographic receipts
Connect stored evidence through BLAKE3 hashes and traceable receipt references.
Explore capabilityAppend-only records
Add new evidence without silently rewriting what was recorded before.
Explore capabilityTenant separation
Keep evidence scoped to the authenticated workspace it belongs to.
Explore capabilityTraceable retrieval
Retrieve evidence and its relationships with non-generative GraphRAG.
Explore capabilityReplay & review
Reconstruct stored memory state and prepare structured evidence exports.
Explore capabilityNo matching capabilities. Try “receipt”, “tenant” or “replay”.
6 capabilities
An evidence-first history.
PHIROK records the source material, retrieved context and configuration associated with an outcome. The record makes these inputs available for later inspection.
- Source and context references stay connected to the record.
- New events extend the history instead of rewriting the past.
- A reviewer can follow the recorded lineage.
A receipt for the record.
Cryptographic receipts connect memory events to recorded data and prior history. BLAKE3 hashes make changes to protected bytes detectable against a trusted reference.
- Inspect receipt, frame and previous-hash references.
- Distinguish evidence integrity from source truth.
- Review the record together with its verification context.
New evidence. Preserved history.
Append-only storage keeps earlier records available as later events are added. Corrections belong in new records with traceable relationships to what came before.
- Append-only write-ahead log (WAL).
- Hash-linked history for tamper evidence.
- Retained events support reconstruction of recorded state.
Evidence stays in its workspace.
Tenant isolation binds access to authenticated backend identity. A record’s tenant scope is part of its context; authorization must be enforced at every boundary.
- Identity comes from authenticated server context.
- Workspace scope is checked for reads and writes.
- A tenant label is not a substitute for access control.
Retrieve the evidence itself.
Non-generative GraphRAG retrieves stored evidence and its relationships. Search results lead back to records a person or agent can inspect.
- Retrieve evidence without generating a new answer.
- Preserve provenance through search and inspection.
- The current semantic layer uses deterministic test fixtures; commercial production embeddings are not advertised.
Return to a recorded state.
Deterministic replay reconstructs stored memory state from retained records. Structured exports support investigation and review.
- Inspect source, context, policy and receipt references.
- Compare recorded decisions and their evidence.
- Replay does not promise identical future LLM responses.
YOUR NEXT DECISION
Build with memory.
Stand behind the evidence.
Give consequential AI a history you can inspect.