# Eigen Mesh

> Your memory. Not theirs.

Eigen Mesh is open-source personal AI memory infrastructure. You capture thoughts in natural language, and Eigen Mesh automatically enriches, classifies, and organizes them into a searchable knowledge base -- accessible to any AI assistant through MCP (Model Context Protocol).

## What it does

- **Capture:** Submit raw thoughts via web UI, voice dictation, or API -- no tagging or structuring required
- **Enrichment:** Automatic classification, entity extraction, embedding, and knowledge graph construction
- **Retrieval:** Hybrid search across vector similarity (pgvector), full-text search, and precomputed graph neighbors and community bundles (no live graph traversal at query time)
- **Grounded answers:** In-app chat and MCP clients retrieve relevant memories, then compose answers with citations
- **Consolidation:** Nightly background jobs for salience decay, ontology pruning, entity deduplication, and community detection

## Core values

- **Open source** -- Apache 2.0 license. Self-host or use managed hosting.
- **Coffee-priced subscription** -- 100 free credits on signup (~100–150 thoughts), then €3.99/mo on managed hosting. Cancel anytime.
- **Self-host free** -- Bring your own API keys; you pay your provider.
- **Data sovereignty** -- Your data stays in your database. Encrypted at rest, tenant-isolated.
- **No vendor lock-in** -- Works across AI tools via MCP. Your memory follows you.

## When to use Eigen Mesh

Eigen Mesh is a good fit when the job is **remembering**, not document search:

- **Personal AI memory across tools.** You use several AI assistants (Cursor, Claude, ChatGPT, or any MCP client) and want one memory that follows you between them instead of being siloed in each tool.
- **Everyday personal facts and tasks.** Preferences, projects, deadlines, people, decisions, recurring questions -- things you tell an assistant once and want it to actually remember. You capture in natural language; no tagging or structuring required.
- **A self-hosted second brain.** You want the same memory you can run on your own hardware (Docker Compose, PostgreSQL + pgvector, Apache 2.0), with your data in your database and your API keys.
- **A team or company brain.** Eigen Enterprise puts your team's projects, timelines, and knowledge in one findable place. Early access -- [contact us](mailto:enterprise@eigenmesh.xyz).

### When Eigen Mesh is NOT a fit

- **Enterprise RAG over large document collections.** Eigen Mesh works on deliberately captured thoughts, not bulk ingestion of PDFs, wikis, or ticket archives. If you need retrieval over thousands of documents, use a RAG pipeline.
- **Real-time collaborative note editing.** Thoughts are personal memory entries with hybrid search, not shared documents with multiplayer editing, presence, or permissions.
- **A chatbot or a model.** Eigen Mesh is the memory layer. It is not an assistant itself.

### How agents interact

Agents and AI assistants connect through four HTTP MCP tools:

- `capture_thought` -- Store a new memory
- `retrieve_thoughts` -- Hybrid search, or browse recent thoughts with `order=created_at`
- `edit_thought` -- Update stored memories
- `delete_thought` -- Archive (soft-remove) a thought

See the MCP documentation for endpoints, authentication, and payloads: [MCP overview](https://eigenmesh.xyz/developers/mcp-overview) ([markdown](https://eigenmesh.xyz/developers/mcp-overview.md)) and [MCP tools](https://eigenmesh.xyz/developers/mcp-tools) ([markdown](https://eigenmesh.xyz/developers/mcp-tools.md)).

## For developers

Eigen Mesh exposes four HTTP MCP tools for AI assistants:

- `capture_thought` -- Store a new memory
- `retrieve_thoughts` -- Hybrid search, or browse recent thoughts with `order=created_at`
- `edit_thought` -- Update stored memories
- `delete_thought` -- Archive (soft-remove) a thought

Notes tools (`create_text_file`, `search_text_files`, etc.) are available in the in-app chat agent only, not over HTTP MCP.

## Tech stack

- **Frontend + API:** SvelteKit with adapter-node
- **Database:** PostgreSQL 16 with pgvector (vector search) and Apache AGE (knowledge graph at ingest and visualization)
- **Auth:** Better Auth with Row-Level Security
- **ORM:** Drizzle
- **Deployment:** Docker Compose (app on port 3000, database on port 5432)

## Navigation

- [Developers docs index](https://eigenmesh.xyz/developers/llms.txt)
- [Documentation](https://eigenmesh.xyz/developers)
- [Full docs export](https://eigenmesh.xyz/developers/llms-full.txt)
- [OpenAPI spec](https://eigenmesh.xyz/openapi.json)
- [Blog](https://eigenmesh.xyz/blog)
- [Pricing](https://eigenmesh.xyz/#pricingSectionTarget)
- [About](https://eigenmesh.xyz/about)
- [Contact](https://eigenmesh.xyz/contact)
- [GitHub repo](https://github.com/hikaru90/eigen)
- [Privacy](https://eigenmesh.xyz/privacy)
- [Imprint](https://eigenmesh.xyz/imprint)
- [Sitemap](https://eigenmesh.xyz/sitemap.xml)

## Links

- Website: https://eigenmesh.xyz
- App: https://app.eigenmesh.xyz
- GitHub: https://github.com/hikaru90/eigen
- License: Apache 2.0

## Self-hosting

```sh
git clone https://github.com/hikaru90/eigen.git && cd eigen
./install.sh --origin https://your-domain --non-interactive
docker compose up -d --build
```

Requires Docker with compose plugin. No Node.js needed at deploy time.