Your AI tools share
the same memory.
One MCP server connects Claude Desktop, Cursor, Windsurf, and every other AI client to structured, persistent memory that follows you across tools and machines.
No re-explaining your project. No losing context when you switch models. Explicit, controlled, encrypted — built the way infrastructure should be.
Works with your tools
How It Works
One MCP server. Your context, available to every tool you use.
Connect a Client
Add Unimatrix to Claude Desktop (via local bridge), Cursor/Windsurf (direct HTTP), or any MCP-compatible tool using your API key. Takes 2 minutes.
Store What Matters
During conversations, the AI (or you) explicitly calls tools like unimatrix_store_memory to save important context into structured Palaces and Locations.
Load Context Deliberately
In your system prompt, tell each client to call unimatrix_list_palaces + search tools at session start. You control exactly when context loads.
Note: You must add explicit instructions in your LLM settings so it loads memory at the start of sessions. No automatic background behavior — you stay in control.
Developers who ship with multiple AIs every day
Real usage patterns from teams running Claude, Cursor, and custom agents together.
I switch between Claude Desktop and Cursor constantly. Before Unimatrix I'd spend the first 5 minutes of every session re-explaining my codebase. Now I just start working.
Marcus T.
Senior Full-Stack Engineer · Remote startup
The Memory Palace model is the right abstraction. It's not a flat key-value dump — it's organized like how I actually think about my projects. The semantic search actually works.
Priya S.
AI tooling developer · Independent contractor
We run 20+ agents concurrently. The spend caps and human-in-the-loop controls are not optional for us — they're table stakes. Unimatrix is the only MCP layer that has them built in.
Daniel K.
ML Infrastructure Lead · Series A SaaS
Native clients for every platform
Desktop apps for macOS, Windows, and Linux — plus mobile companions — all sharing the same cloud MCP memory backend.
Desktop
No download needed
Use the Web App
Browse memories, manage workspaces, and configure your API keys from any browser — no install required.
Open Web App →The Collaborative AI Room
Ask one question. Every AI in the room answers independently — then they listen to each other.
Claude spots a hallucination in GPT's answer. Gemini adds context the others missed. Grok pushes back on the consensus. The room works as a team — no auto-routing, no silent delegation. Every model speaks, and the best ideas survive.
Ask the Room
You post one question or task. Every connected AI model receives it simultaneously — no model is pre-selected as the lead.
Independent Answers
Each LLM responds on its own. Different models catch different things — one may hallucinate where another is precise.
Collaborative Refinement
Models can see each other's answers and respond: correct a mistake, add missing context, or suggest a better direction. Teamwork, not just parallelism.
No credit card needed for the trial. Available on Pro and Enterprise plans.
Built for the multi-AI world
You use multiple AIs. Unimatrix makes them feel like one.
Cross-Client Memory
The same memories are readable and writable from Claude Desktop, Cursor, Windsurf, custom agents, or your own scripts via MCP.
Memory Palace Structure
Memories live in hierarchical Palaces and Locations (method of loci model), not a flat key-value bag. Better for agents, better for you.
Self-Host or Cloud
Use our managed cloud or run the full stack yourself with Docker + PostgreSQL + pgvector for complete data sovereignty.
Standard MCP Interface
Works with any client that speaks the Model Context Protocol. No proprietary SDKs. No per-model glue code.
Explicit Control
You decide when tools are called. No hidden background syncing, no surprise data exfiltration. You are always in control.
Full Tool Schemas
unimatrix_list_palaces, unimatrix_store_memory, unimatrix_search_memories. Complete schemas via tools/list.
Agent Spend Limits
Per-agent daily spend caps, real-time cost tracking, and automatic budget alerts. Never get surprised by runaway agent costs.
Human-in-the-Loop
Configurable approval gates for sensitive actions. Agents can request permission before executing high-impact tools.
Full Audit Trail
Detailed token usage logs, cost attribution per agent, and complete audit trail for every approval and configuration change.
Production-Grade Agent Controls
Run dozens of agents safely with real financial guardrails and human oversight.
Spend Limits & Budgets
Set daily caps per agent. Get alerts at 80%. Hard blocks when limits are reached. Never get an unexpected invoice.
Human-in-the-Loop
Fine-grained approval workflows. Sensitive tools can require explicit human sign-off before execution — for every agent.
Full Observability
Per-agent token usage, cost attribution, anomaly detection, and complete audit logs for compliance and post-mortems.
Unimatrix vs. Mem0
Unimatrix and Mem0 solve different problems.
Mem0 is a lightweight, embeddable memory layer designed primarily for single-agent applications. You typically run it yourself as a key-value store inside your own agent loop.
Unimatrix is a managed, multi-tenant MCP server built for developers who use many different AI clients (Claude Desktop, Cursor, Windsurf, Continue, custom agents). It provides a standardized protocol interface, hierarchical Memory Palace organization, semantic search, and first-class cross-client federation — without writing per-tool glue code.
Choose Mem0 if you want a simple store inside one agent you fully control.
Choose Unimatrix if you want your context available to every MCP client you already use.
Security & Architecture
Built as infrastructure. Not magic.
Encryption & Access
- •Memories encrypted at rest with AES-256-GCM (scrypt KDF per user)
- •TLS 1.3 in transit — everywhere
- •Clerk JWT authentication + PostgreSQL Row Level Security
- •Raw API keys are never stored — only bcrypt-hashed prefixes
Data Control & Transparency
- •Full delete and export via API at any time
- •Your data is never used for training
- •All MCP tool calls are auditable per account
- •Self-host the full stack: Docker + Postgres + pgvector
Full details in SECURITY.md and the self-host guide.
Simple Pricing
Start free. Scale when you need it. Cancel anytime.
Free
No credit card required
- ✓3 memory workspaces
- ✓200 memories
- ✓2 devices
- ✓Full-text & semantic search
- ✓All LLMs via MCP + REST
- ✓End-to-end encrypted
Pro
or $79 / year — save $29
- ✓Unlimited workspaces
- ✓Unlimited memories
- ✓Unlimited devices
- ✓20 API keys
- ✓Collaborative AI Room
- ✓Agent Spend Controls & HITL
- ✓Full Audit Logs & Telemetry
- ✓Priority support
- ✓14-day free trial
Enterprise
or $299 / year — save $49
- ✓Everything in Pro
- ✓Self-hosted Docker stack
- ✓Team memory sharing
- ✓Agentic device control
- ✓100 API keys
- ✓SSO / SAML
- ✓SLA + dedicated support
All paid plans include a 14-day free trial. Cancel anytime. Payments secured by Stripe.
Frequently Asked Questions
How is Unimatrix different from ChatGPT memory?
ChatGPT memory is siloed inside ChatGPT. Unimatrix is a protocol layer — your memories are available to Claude, Cursor, Windsurf, and any other MCP-compatible tool simultaneously. You own the data and control exactly when it's accessed.
Does it work automatically or do I have to configure it?
Intentionally explicit. You add a system prompt to each AI tool instructing it to load your memory at the start of sessions. No hidden background syncing — you always know exactly what context each AI has. This is how MCP memory actually works.
What counts as a 'memory'?
Any piece of text you store — a note, a fact, a code snippet, a meeting summary, a project decision. Each individual entry is one memory. They live inside named Palaces (workspaces), organized into Locations.
Is my data encrypted?
Yes. Data is encrypted at rest with AES-256-GCM. TLS 1.3 in transit. API keys are bcrypt-hashed — we never see the raw value. Payments are processed by Stripe — we never touch your card details.
Can I self-host the entire thing?
Yes — the full stack (server, web app, worker) runs on Docker + PostgreSQL + pgvector. Enterprise plan includes the self-host guide and dedicated setup support.
Can I cancel anytime?
Yes — cancel from Settings at any time. You keep full Pro access until the end of your billing period. No questions asked.
Not ready to sign up yet?
Get notified when we ship new MCP integrations, desktop app updates, and enterprise features. No spam. Unsubscribe anytime.
Stop re-explaining yourself to every AI.
Your context belongs to you — not to a single chat window on a single device. Unimatrix makes every AI smarter by giving it your full history, everywhere.
Free tier · No credit card · Cancel anytime