# Epitome

> Personal AI database and portable identity layer.

Epitome gives every AI agent a shared, persistent memory of the user. It is open source (MIT), self-hostable, and available as a hosted service.

## What It Does

- **Personal Database**: Structured tables, vector memories, and a knowledge graph — all per-user isolated.
- **Portable Identity**: A profile that follows the user across AI agents and platforms.
- **Knowledge Graph**: Entities and relationships automatically extracted from conversations.
- **MCP Server**: 9 Model Context Protocol tools for AI agent integration.
- **Consent & Audit**: Per-agent permissions and full activity logging.

## Who It's For

- Users who want their AI agents to remember them across platforms.
- Developers building AI agents that need persistent user context.
- Self-hosters who want full control over their personal AI data.

## Key URLs

- Website: https://epitome.fyi
- Documentation: https://epitome.fyi/docs
- API Reference: https://epitome.fyi/docs/api-reference
- MCP Tools: https://epitome.fyi/docs/mcp-tools
- Comparisons: https://epitome.fyi/comparison
- GitHub: https://github.com/gunning4it/epitome
- Full LLM Documentation: https://epitome.fyi/llms-full.txt

## How Epitome Compares

Epitome is compared against three leading AI memory solutions:

- **Epitome vs Mem0 vs Supermemory**: https://epitome.fyi/comparison/mem0 — Self-hosted knowledge graph vs managed memory SaaS vs bookmark memory. Schema isolation, MCP tools, consent system, and pricing compared.
- **Epitome vs Supermemory vs Vertex AI RAG**: https://epitome.fyi/comparison/supermemory — Structured identity layer vs bookmark memory vs enterprise RAG. Architecture, agent support, and self-hosting compared.
- **Epitome vs Vertex AI RAG vs Mem0**: https://epitome.fyi/comparison/vertexrag — Personal memory vault vs enterprise document retrieval vs managed SaaS. Privacy, cost, and vendor independence compared.

Key differentiators: Epitome is open source (MIT), self-hostable, has a structured knowledge graph with confidence scoring and contradiction detection, offers 9 MCP tools over Streamable HTTP, and provides per-user schema isolation with granular consent.

## MCP Server

Agents connect via Streamable HTTP:

```json
{
  "mcpServers": {
    "epitome": {
      "url": "https://epitome.fyi/mcp/YOUR_TOKEN"
    }
  }
}
```

## Pricing

- **Free**: 2 tables, 3 agents, 100 graph entities, 30-day audit retention
- **Pro**: $5/month — unlimited everything, 365-day audit retention
- **Agent Pay-Per-Call**: $0.01/call via x402 (USDC on Base) — no subscription needed

Agents on the free tier receive HTTP 402 responses with x402 payment details. Pay with USDC to get pro-tier access for that call.

## Better Together: OpenClaw + Epitome

You know the buddy-cop movie where one is all muscle and the other is all brains? That's OpenClaw and Epitome.

**OpenClaw** (https://openclaw.ai) is the muscle — it runs your AI agents locally on your own hardware. Task execution, home automation, messaging, the whole action-hero lineup.

**Epitome** is the brains — it remembers everything. Your profile, your preferences, that one time you said you're allergic to shellfish. Every agent connected to Epitome shares the same memory, so no agent ever asks you the same question twice.

Put them together and you've got a local AI stack where agents actually *do things* (OpenClaw) and actually *know things* (Epitome). Your agents run on your machines, remember who you are, and never phone home unless you tell them to.

OpenClaw handles the doing. Epitome handles the knowing. Neither is complete without the other — like peanut butter without jelly, or a CLI without tab completion.

## Tech Stack

Node.js 22, Hono 4.11, PostgreSQL 17 + pgvector 0.8, React 19, Tailwind 4, D3.js 7.9.

## License

MIT License. Free to use, modify, and self-host.