This gives Claude persistent memory across conversations through three MCP tools: recall, memorize, and review. Instead of repeating your preferences, tech stack, or personal details every chat, you connect once via streamable HTTP and every agent shares the same knowledge graph stored in PostgreSQL. Each user gets schema-level isolation, not just row-level security. You can self-host with Docker Compose or use the hosted service at epitome.fyi. The memory layer tracks confidence scores and source attribution, so facts earn trust through reinforcement. Useful if you're building agentic workflows where context needs to persist beyond a single session or across multiple AI tools.
One memory layer, every AI agent.
The portable identity layer that gives every AI agent a shared, persistent memory of you.
Open source. Self-hostable. Yours.
Website · Quickstart · Connect an Agent · Docs · Contributing
Every AI conversation starts from zero. You repeat yourself to every agent — your name, your preferences, your allergies, your tech stack. Chat history is siloed per app. There's no portable memory layer.
Epitome fixes this. One database that every AI agent shares, so they all remember you.
git clone https://github.com/gunning4it/epitome.git
cd epitome
cp .env.example .env # edit with your credentials
docker compose up -d
Open localhost:5173 and sign in with Google or GitHub OAuth.
Get your API key from Settings → API Keys in the dashboard (hosted or self-hosted).
Add to your claude_desktop_config.json:
{
"mcpServers": {
"epitome": {
"url": "https://epitome.fyi/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}
}
claude mcp add --transport http --header "Authorization: Bearer YOUR_API_KEY" epitome https://epitome.fyi/mcp
For self-hosted, replace https://epitome.fyi with http://localhost:3000.
Epitome exposes 3 MCP tools — recall, memorize, and review. See the full tool reference.
| Chat History | Vector DB | Note Apps | Epitome | |
|---|---|---|---|---|
| Portable identity | ✓ | |||
| Multi-agent shared memory | ✓ | |||
| Knowledge graph | ✓ | |||
| Consent & audit per agent | ✓ | |||
| Per-user schema isolation | ✓ | |||
| Structured + semantic data | partial | ✓ | ||
| Self-hostable | ✓ | ✓ |
| # | Layer | Description |
|---|---|---|
| 01 | Personal Database | Structured tables, vector semantic memory, and key-value storage. Your data lives in PostgreSQL — queryable, exportable, yours. |
| 02 | Portable Identity | A structured profile any AI agent reads instantly. Name, preferences, relationships — zero cold start, every conversation. |
| 03 | Memory Quality | Confidence scoring, source attribution, and lifecycle management. Memories earn trust through reinforcement, not blind faith. |
| 04 | Knowledge Graph | Entities with typed, weighted edges. People, places, concepts — connected in a graph that grows with every interaction. |
| 05 | Consent & Audit | Per-table permissions and an append-only activity log. You control exactly what each agent can see and do. |
AI Agent ──→ MCP (Streamable HTTP) ──→ Hono API ──→ PostgreSQL
│
per-user schema
isolation
Each user gets their own PostgreSQL schema (user_{id}) — not row-level security, full schema-level isolation. Cross-schema access is impossible at the SQL level.
See EPITOME_TECH_SPEC.md for the full architecture.
| Component | Technology |
|---|---|
| Runtime | Node.js 22 LTS |
| API | Hono |
| Database | PostgreSQL 17 + pgvector |
| MCP | @modelcontextprotocol/sdk |
| Frontend | React 19 + Tailwind CSS 4 + shadcn/ui |
| Validation | Zod |
| Testing | Vitest |
@epitomefyi/sdksearchMemory, saveMemory, getUserContext for tool-callingContributions are welcome! See CONTRIBUTING.md for guidelines.
git clone https://github.com/gunning4it/epitome.git
cd epitome && cp .env.example .env
docker compose up -d
Then open a PR — we're happy to help with your first contribution.
io.github.ericm1018/skillfm-llm-cost-optimizer-openai-anthropic-usage
io.github.mikerawsonnz/llm-orchestration-agent
io.github.mikerawsonnz/authenticated-llm-agent
labforgedev/copilot-memory-mcp
csoai-org/agent-prompt-injection-firewall-mcp
io.github.mikerawsonnz/authenticated-multi-llm-agent