UNIMATRIX / PERSISTENT AI MEMORY
What you know
stays with you.
Unimatrix gives your AI tools a shared memory for the conversations, decisions, and projects that matter. Bring context forward through explicit, auditable connections.
Memory formation is a pipeline, not a raw transcript dump.
The Librarian turns explicit inputs into durable, searchable context, preserving structure and provenance for later recall.
NORMALIZATION
Clean incoming tool calls, extension saves, imports, and API writes into a consistent memory envelope.
DECOMPOSITION
Split long conversations, files, and notes into atomic memory candidates with source context intact.
EXTRACTION
Extract summaries, entities, semantic triples, tags, embeddings, and searchable full-text signals.
CONTRADICTION
Compare new claims against existing beliefs and preserve supersession lineage when facts change.
PERSISTENCE
Persist durable memory into Spaces, Locations, and Memories backed by PostgreSQL, pgvector, and tsvector.
A memory substrate with a clear data path.
Public clients connect to the Next.js web endpoint. Internal MCP and worker services perform deeper processing against the canonical PostgreSQL store, combining pgvector similarity with full-text retrieval inside the same governed data path.
Spaces → Locations → Memories
Unimatrix uses a canonical hierarchical structure to isolate context scopes cleanly instead of relying on flat log blobs.
{
"space": {
"id": "spc_work_9a2f",
"name": "Work Architecture",
"parent_id": null,
"locations": [...]
}
}An example workspace: two spaces, each holding locations, each holding individual memories. Example content, not account data.
Read this structure as text
- Work
- Auth System
- Custom JWT middleware, not the standard library
- Signing keys rotate quarterly
- Billing
- Stripe webhooks must stay idempotent
- Annual plans prorate on upgrade
- Auth System
- Research
- Sources
- Cite primary sources by DOI
- Skip vendor blogs for benchmarks
- Drafts
- Open with the reader problem, not the method
- Section three still needs evidence
- Sources
A memory should be able to explain where it came from.
Veritas tracks supersession and contradiction history so updated facts can replace stale ones without erasing the audit path. The lineage tools expose provenance for critical memory decisions.
The production datastore is PostgreSQL with pgvector.
A migration note supersedes stale Neon-specific language.
The current answer can explain which memory won and what it replaced.
MCP, REST, extensions, mobile, and Loom.
Unimatrix supports multiple explicit capture and recall surfaces. Loom is represented as guided ingest, preview, status, and approval tooling rather than a blanket promise to absorb every source automatically.
| Surface | Connection | Status |
|---|---|---|
| Claude Desktop | Streamable HTTP MCP | Supported |
| Cursor / Windsurf / Continue | MCP configuration | Supported |
| ChatGPT, Claude.ai, Gemini, Grok | Manifest V3 extension save flow | Supported where extension is installed |
| Custom agents and backends | REST tools and /api/tools/call | Supported |
| Mobile app | React Native API client | Supported app surface |
| Loom ingest | Guidance, ingest, preview, approve tools | Available through dashboard/API surfaces |
Connect an AI client in minutes.
The fastest path is explicit: generate a token, configure the client, and instruct the assistant to recall or store memory when appropriate.
Open quickstartCommercial Plans & Economics
Transparent pricing built for individuals, teams, and enterprise infrastructure.
For individual experimentation
- ✓3 spaces & 200 memories
- ✓1 API key
- ✓MCP & extension access
- ✓Full-text & vector retrieval
- ✓Community support & docs
Billed annually ($79/yr)
- ✓Unlimited spaces & memories
- ✓20 API keys & unlimited devices
- ✓Cross-client memory continuity
- ✓Full per-account audit logs
- ✓Early access to new features
Per user/year (Min 5 users)
- ✓Everything in Pro
- ✓50 API keys for team agents
- ✓Organization RBAC & RLS isolation
- ✓Human-in-the-loop (HITL) approval
- ✓Veritas review where enabled
For organizational memory infrastructure
- ✓Self-host deployment (Docker + Postgres)
- ✓Private / Dedicated Enterprise VPC option
- ✓Audit review and deployment support
- ✓Priority SLA & custom onboarding
- ✓Security & procurement support
Bring persistent memory to your AI stack
Stop letting AI context evaporate between chat windows. Connect your AI clients to a single, governed persistent memory layer.