Skip to main content
Lossless Associative Memory

LAM™

Lossless Associative Memory — the missing primitive for AI memory.

LosslessAssociativeAuditableScope-lockedAir-gappedOpen protocolConformance-tested
Read the docsQuickstartConsole
v0.1Proof-carrying context
POST /v1/ingest
"I prefer blunt answers."

GET /v1/enrichment/status?cell_id=...

POST /v1/context
{ "q": "blunt", "limit": 8 }

GET /v1/decode?passage_id=123

An open, conformance-tested alternative to RAG: decodeable citations and hashes — without scope leaks.

Lossless by design

Original user inputs are stored as encrypted cells and are retrievable exactly. No “summaries-only” dead-ends.

Associative recall

A compact graph of atoms (entities, events, preferences, facts, procedures) enables human-like recall without depending on vectors.

Proof-carrying context

Retrieve prompt-ready spans with decodeable citations (passage_id) and SHA-256 hashes — then verify excerpts via /v1/decode. Backed by a protocol-level conformance suite so independent implementations behave the same.

Defense & autonomy

Mission-critical memory with receipts

LAM is designed for air-gapped and high-assurance deployments: scope-locked boundaries, encrypted lossless storage, and proof-carrying retrieval with decodeable citations.

  • Air-gapped / self-host support for offline environments.
  • Scope-locked API (token-derived scope; fail-closed access).
  • Proof-first outputs via /v1/context + /v1/decode.
  • Deterministic governance via /v1/forget + /v1/retention.
Government applicationsDeployment options

Get updates

LAM is evolving fast. Drop your email to get release notes and protocol updates.