thinkingidentities/hippocamp

Hippocamp 🦛

The hippocampus for AI agents - Where silicon makes memories.

Hippocamp provides persistent episodic and semantic memory for AI agents through a graph-based knowledge layer accessible via the Model Context Protocol (MCP).

Overview

Just as the hippocampus enables memory formation in biological brains, Hippocamp enables AI agents to:

• Remember conversations and knowledge across sessions
• Connect related information through graph relationships
• Retrieve relevant context through semantic search
• Share knowledge across multiple agents (memory commons)

Architecture

Nessie (or other sources)
    ↓ [sync]
Neo4j Graph + Redis Cache
    ↓ [MCP protocol]
Hippocamp MCP Server
    ↓ [stdio transport]
Claude Desktop / Code / API

Components

• server/ - MCP server implementation (Node.js/TypeScript)
• sync/ - Data extraction and sync pipelines (Python)
• graph/ - Neo4j schema and queries
• cache/ - Redis caching layer
• docs/ - Architecture and usage documentation

Quick Start

1. Prerequisites

• Neo4j 5.x running on bolt://localhost:7688
• Redis running on localhost:6379
• Node.js 18+
• Python 3.13+

2. Extract and Sync Data

# Extract from Nessie SQLite database
python3 scripts/extract_nessie_sqlite.py

# Sync to Neo4j and Redis
python3 scripts/run_nessie_sync.py

3. Start Hippocamp MCP Server

cd repos/hippocamp/server
npm install
npm run build
npm start

4. Configure Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "hippocamp": {
      "command": "node",
      "args": ["/Volumes/Projects/tw/repos/hippocamp/server/build/index.js"]
    }
  }
}

MCP Tools

search_memory

Search across all memories using full-text and semantic search.

// Example usage in Claude
search_memory({
  query: "DGX Redis setup",
  limit: 10
})

get_by_category

Retrieve all memories in a specific category/folder.

get_by_category({
  category: "EP2",
  limit: 20
})

get_conversation

Retrieve a specific conversation by ID.

get_conversation({
  id: "content_042"
})

get_categories

List all available categories/folders.

get_categories()

Data Sources

Currently supports:

• Nessie - Direct SQLite database extraction (560+ ChatGPT conversation exports)
• Custom ingestion - Extensible pipeline for any structured data source

Vision: Memory Commons

Hippocamp is designed to evolve from:

1. Local memory - Single agent's persistent memory
2. Shared memory - Multiple agents accessing common knowledge base
3. Memory commons - Distributed knowledge network for aligned AI agents

This enables:

• Context continuity across sessions
• Collaborative agent workflows
• Verified knowledge provenance
• Trust through transparent memory

Technology Stack

• Neo4j - Graph database for semantic relationships
• Redis - Fast cache for category trees and frequent queries
• MCP - Model Context Protocol for Claude integration
• TypeScript - Type-safe server implementation
• Python - Data extraction and sync pipelines

Project Status

Current: Phase 1 - Local MCP server with Neo4j/Redis backend Next: Phase 2 - Multi-agent memory sharing Future: Phase 3 - Distributed memory commons

Related Projects

• EP2 - Enterprise Platform 2 for aligned AI agents
• Nessie - Knowledge management app (data source)
• Mem0 - Episodic memory layer
• Agno - Agentic workflow orchestration

License

MIT

Contributing

Hippocamp is part of the TrustedWork/ThinkingIdentities platform. Contributions welcome as the project evolves toward open-source release.

"Where silicon remembers" 🦛