Hierarchical memory model
Palaces → Locations → Memories keep long-running context organized instead of storing everything in one flat log.
Persistent memory for every AI you use, every day
One memory layer.Every AI surface.
Capture memory the way you already work — MCP in Claude Desktop and your IDE, tool calls from your own agents, a browser extension on ChatGPT, Claude.ai, Gemini, and Copilot, or a direct API sync. It all lands in one memory layer that survives model switches, device switches, and everyday life.
✓ Explicit memory loading✓ Cloud + self-host paths✓ Stripe-secured subscriptions
Stop being your AI's context router
You don't need to understand MCP internals to see the problem: every new chat window starts from zero, and you're the one stitching the context back together by hand — whether that's a class project, a novel, a research thread, or a codebase.
Without Unimatrix
⟳
Claude Desktop
Re-paste project context, decisions, and conventions — again
⟳
Cursor
Re-paste project context, decisions, and conventions — again
⟳
ChatGPT
Re-paste project context, decisions, and conventions — again
Same explanations copy-pasted into three different windows. Context drifts between them, and decisions quietly get forgotten.
With Unimatrix
One memory layer
Palaces → Locations → Memories
✓ Claude Desktop
✓ Cursor
✓ ChatGPT
Store it once. Every connected client reads from the same memory layer — no re-explaining, no context drift, completely synchronized facts.
Platform capabilities built for durable memory systems
Built on a modern memory architecture that's simple enough for daily use and solid enough for the teams and power users who depend on it every day.
Every capture channel Unimatrix supports, converging on a single memory layer.
- MCPClaude Desktop and IDE agents read and write memory over the MCP server.Claude Desktop · Cursor · Windsurf · VS Code
- Browser extensionCaptures context straight from web LLMs that have no MCP support.ChatGPT · Claude.ai · Gemini · Copilot
- Agent tool callsYour own agents call remember and recall as ordinary tools.Custom agents · Automation workflows
- REST APIAnything that can make an HTTP request can sync memory directly.Mobile app · Your backend
Four ways in, one memory layer
MCP for Claude Desktop and IDE agents, tool calls from your own agents, a browser extension for ChatGPT, Claude.ai, Gemini and Copilot, and a REST API — all reading and writing the same memories.
Retrieval engineered for production
Semantic + full-text search, confidence scoring, contradiction detection, and session-aware retrieval keep context relevant.
Real-time sync and observability
Ably-powered updates, full audit logs, RBAC controls, and approval gates keep everyday use safe — for individuals, power users, and teams alike.
Via MCP
Claude DesktopCursorWindsurfContinueVS Code MCPWarp
Via browser extension
ChatGPTClaude.aiGeminiCopilotPerplexityGrok
Via tool calls & REST API
Custom agentsAutomation workflowsMobile appYour own backend
Explore the memory hierarchy
Unimatrix divides information into logical containers. Click on the nodes below to see how Unimatrix organizes facts and exposes them via clean schemas.
Memory payload output
JSON output
{
"palace": {
"id": "plc_work_9a2f",
"name": "Work Workspace",
"locations": [...]
}
}Any MCP client or API integration can query this structure instantly using standard tools like unimatrix_get_palace.
An example workspace: two palaces, 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
Security, governance, and deployment parity
Unimatrix is built as infrastructure, not a toy feature. Client-side encryption ensures we cannot read your data, while managed and Dockerized self-hosted modes match corporate compliance rules.
Managed Cloud
Postgres + pgvector instances automatically deployed, backed up, and optimized for speed. Best for quick startups.
Self-Hosted
Full data residency. Deploy the entire memory database locally using Docker Compose, preserving security inside private networks.
●AES-256-GCM encrypted memory at rest
●TLS 1.3 for all transport
●bcrypt-hashed API keys
●Per-account audit logging
●Managed cloud hosting with Postgres + pgvector
●Self-host option with Docker + PostgreSQL
Stripe billing, clearly presented
Choose monthly or annual billing. Apply promotion codes at checkout and manage your subscription securely through Stripe.
Monthly BillingAnnual BillingSave 25%
Free
$0
For anyone getting started with AI memory
- ✓3 workspaces
- ✓200 memories
- ✓1 API key
- ✓2 devices
- ✓MCP, extension, and REST access
- ✓Ad-supported
Pro
$9 / month/ month
For anyone who relies on AI every day
- ✓Unlimited workspaces and memories
- ✓20 API keys
- ✓Cross-client memory workflows
- ✓Audit logs and usage visibility
- ✓Stripe monthly or annual billing
- ✓No ads
Enterprise
$29 / month/ month
For teams and regulated environments
- ✓Everything in Pro
- ✓100 API keys
- ✓Self-host deployment
- ✓Advanced governance controls
- ✓Priority support path
- ✓No ads
Billing Policy:Manage subscriptions, invoices, and updates securely. You can cancel or upgrade directly via the Stripe customer portal inside your Account settings at any time.
Stop re-explaining yourself to every AI
Set up one memory layer and keep context portable across sessions, tools, and devices — no matter how you use AI, every day.