Enigma
The memory card for AI

Don’t trust it. Check it.

Enigma gives people and teams a provider-neutral Memory Passport: portable context, local-first custody, scoped model injection, and signed receipts for Enigma-mediated memory events — verifiable offline, no server required.

Receipts verify declared Enigma-mediated operations. They do not prove factual truth, model forgetting, complete provider deletion, or uninstrumented side channels.

Scroll
The AI memory problem

Memory is the
new lock-in.

The most useful AI remembers your preferences, projects, and decisions — but that memory is scattered across chat histories, provider memory, vector stores, agent logs, and browser state. Switch tools and you lose continuity. Ask for an audit trail and you get a dashboard.

  • Provider lock-inMemory stays trapped inside one subscription, model, or agent stack.
  • Audit gapLogs and traces aren’t portable cryptographic receipts.
  • Governance gapTeams can’t show what was retrieved, injected, denied, or tombstoned.
  • Deletion ambiguity“Deleted” rarely comes with evidence anyone can check.
The memory-card move for AI

Carry it
like a passport.

A memory card didn’t need to own every camera to make storage portable. Enigma doesn’t need to own every assistant to make memory portable: a canonical vault under your custody, scoped context into each model call, and a verifiable trail of what the memory layer did.

  • 01Memory Passport

    Portable AI context that moves across clients, agents, and workflows.

  • 02Local-first vault

    Canonical memory in a local bundle, or in customer-controlled infrastructure for enterprise.

  • 03Scoped context

    Send a model the context for the task — not your entire memory system.

  • 04Signed receipts

    Verify create, retrieve, pack, boundary, export, and tombstone events offline.

From memory to proof

Vault → policy → context → receipt.

  1. VaultCreate or import

    Write memories through the CLI, MCP server, or importers into a local-first vault. Provider exports become canonical only once Enigma writes them.

  2. PolicyRetrieve scoped context

    An MCP-capable agent asks Enigma for context. Policy decides what’s allowed; Enigma builds a context pack for the stated purpose.

  3. BoundaryUse a model or tool

    Scoped context crosses the model/tool boundary. Provider-native memory is treated as cache — Enigma stays the system of record.

  4. ReceiptVerify offline

    Export a proof bundle and run the verifier. A valid receipt proves the operation was signed, ordered, and committed under a stated policy.

The receipts

Every operation
leaves a proof.

A value-blind receipt anyone can verify offline — no server, no trust required. One zero-dependency library covers the lifecycle.

  • Custody
  • Grounding
  • Completeness
  • Consent
  • Data-minimization
  • Forgetting
Forgetting-Closure

Prove it
forgot.

Tombstone a fact and Enigma proves it is unreachable in the deductive closure of the signed survivors, under a fixed rule-set R — a deletion from Enigma’s serving state you can check, not just trust.

Scoped honestly: unreachable under rule-set R — not a claim of provider-side erasure or model forgetting. Deletion is from Enigma’s local serving state; it does not prove provider-side removal or model-level forgetting.

One memory layer, four paths

Built for users, builders, and teams.

  • UsersCarry your context

    Keep preferences, projects, and style between tools. Install locally, connect your AI client, keep a verifiable trail.

  • DevelopersMemory over MCP

    Add provider-neutral memory through MCP and documented contracts instead of rebuilding vault, policy, and receipt infrastructure.

  • EnterpriseUnder your control

    Customer-controlled custody, gateway decisions, minimized SIEM evidence, and offline-verifiable proof bundles. Hosted/BYOC deployments are waitlist/operator-deploy only; local relay and gateway binaries are bootstrap probes.

  • Network · optionalCoordinate, don’t depend

    Optional relay, witness, and gateway roles for encrypted sync and checkpoints. Local Enigma works without any network — and without a token.

See the trust loop

Create memory. Export proof. Verify offline.

The simplest Enigma demo takes a local vault from empty state to verifiable evidence — no hosted services in the loop.

# install + run the local trust loop
cd enigma
npm install -g .
enigma init --bundle ./.enigma/bundle.json --subject local-user
enigma remember --bundle ./.enigma/bundle.json --text "Prefers concise release notes." --purpose user_memory
enigma context --bundle ./.enigma/bundle.json --query "release notes" --out ./.enigma/context-pack.json
enigma export  --bundle ./.enigma/bundle.json --out ./.enigma/export.json
enigma verify  --bundle ./.enigma/export.json   ✓ receipt valid

The bundle is local. Exported artifacts carry encrypted or committed state and receipt metadata — not raw memory plaintext.

Connects under the tools you already use
  • Claude Desktop
  • Cursor
  • Kimi Code
  • VS Code / Cline
  • Roo Code
  • OpenCode
  • Generic MCP
What it does — and doesn’t

Verifiable doesn’t mean magical.

Enigma is precise about what it proves. Receipts prove facts about Enigma-controlled vault state, receipt chains, context packs, and declared boundary operations — nothing more.

  • Can verify an Enigma vault no longer serves a tombstoned memory address.
  • Can’t verify a closed provider deleted every internal copy, or that a model forgot.
  • Can verify a gateway decision followed a stated Enigma policy hash.
  • Can’t prove facts outside the declared boundary without independent evidence from those systems.
The evidence
0Receipt modules
0Substrate modules
0/104Suites green
0Attacks held

One zero-dependency library, independently red-teamed. Open source. No token.

Carry it with you

Own what
it remembers.

Value-blind facts under classical discrete-log. Not FHE, not post-quantum, not yet audited for deployment. Local Enigma needs no token. © 2026 Enigma · open source · enigmamemory.com