Engram
Engram is a local-first memory server that lets AI agents remember things across sessions with zero cloud dependencies.
README
Engram
Give your AI agents persistent memory. Zero cloud. Zero API keys. Two-minute setup.
Engram is a local-first memory server that lets AI agents like Claude, Cursor, Cline, and Windsurf remember things across sessions. It runs entirely on your machine — no external services, no data leaving your device, no monthly bills.
npm install -g @hbarefoot/engram
engram start
That's it. Your AI agent now has long-term memory.
Why Engram?
Most AI memory solutions require cloud infrastructure, API keys, and external vector databases. Engram takes a different approach:
| Engram | Mem0 | Zep | |
|---|---|---|---|
| Setup time | 2 minutes | Requires API keys + cloud setup | Requires Docker + API keys |
| Dependencies | None — SQLite + local embeddings | OpenAI API, vector DB (Qdrant/Pinecone) | PostgreSQL, OpenAI API |
| Data location | Your machine only | Cloud (or self-hosted with infra) | Cloud (or self-hosted with infra) |
| Cost | Free forever | Pay per API call | Pay per API call |
| Privacy | Complete — nothing leaves your device | Data sent to external APIs | Data sent to external APIs |
| MCP native | ✅ First-class | ❌ REST only | ❌ REST only |
Engram is built for developers who want AI memory without the overhead.
Features
- 🔒 Fully Local — SQLite database + local embeddings (all-MiniLM-L6-v2, 23 MB). No network calls, ever
- 🤖 MCP Native — First-class Model Context Protocol integration. Works with Claude Desktop, Claude Code, Cline, Cursor, Windsurf, and any MCP client
- 🔍 Hybrid Search — Combines vector similarity with full-text search (FTS5) for accurate recall
- 🧹 Smart Deduplication — Automatically detects and merges similar memories (>0.92 similarity threshold)
- 📊 Feedback Loop — Rate memory usefulness to improve future recall accuracy
- 🔐 Secret Detection — Automatically blocks API keys, passwords, and tokens from being stored
- ⏰ Temporal Queries — Filter memories by time: "last week", "3 days ago", or exact dates
- 📦 Namespace Isolation — Organize memories by project, client, or any scope you need
- 🌐 REST API — Full HTTP API with CORS support for custom integrations
- 🖥️ Web Dashboard — React-based UI for browsing, searching, and managing memories
- 💾 Export — Export memories to Markdown, JSON, or plain text for documentation
Quick Start
1. Install
npm install -g @hbarefoot/engram
2. Start the server
engram start
3. Connect to your AI agent
Add Engram to your MCP client config:
Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"engram": {
"command": "engram",
"args": ["start", "--mcp-only"]
}
}
}
Claude Code:
claude mcp add engram -- engram start --mcp-only
Cline / Cursor / Windsurf: Add the same MCP config to your editor's settings. Engram's built-in Integration Wizard can auto-detect your setup:
engram connect
4. Use it
Once connected, your AI agent can store and recall memories naturally:
You: "Remember that our API uses JWT tokens with 24-hour expiry"
Claude: stores the memory via
engram_rememberYou: (next day) "What authentication approach are we using?"
Claude: recalls via
engram_recall— "Your API uses JWT tokens with 24-hour expiry."
Memories persist across sessions, restarts, and even different AI clients sharing the same Engram instance.
MCP Tools
Engram exposes 6 tools to AI agents via the Model Context Protocol:
| Tool | Description |
|---|---|
engram_remember |
Store a memory with category, entity, confidence, namespace, and tags |
engram_recall |
Retrieve relevant memories by semantic query with optional filters |
engram_forget |
Delete a specific memory by ID |
engram_feedback |
Rate a memory as helpful or unhelpful to improve future recall |
engram_context |
Generate a pre-formatted context block (markdown/xml/json) with token budget |
engram_status |
Health check with memory count, model status, and configuration info |
Memory Categories
Memories are organized by type for better retrieval:
- fact — Objective truths about setup, architecture, or configuration
- preference — User likes, dislikes, and style choices
- pattern — Recurring workflows and habits
- decision — Choices made and the reasoning behind them
- outcome — Results of actions taken
CLI Reference
engram start # Start MCP + REST server
engram start --mcp-only # MCP server only (for agent integration)
engram start --port 3838 # Custom port for REST API
engram remember "content" # Store a memory from the command line
engram recall "query" # Search memories
engram forget <id> # Delete a memory
engram list # List all memories
engram status # Health check and stats
engram export # Export memories to JSON
engram import <file> # Import memories from file
engram consolidate # Run deduplication and cleanup
engram agents # List detected AI agents on your system
engram connect # Interactive setup wizard for MCP clients
REST API
The REST API runs on http://localhost:3838 by default.
| Endpoint | Method | Description |
|---|---|---|
/health |
GET | Health check |
/api/status |
GET | System status with stats |
/api/memories |
POST | Create a memory |
/api/memories |
GET | List memories (with pagination, category/namespace filters) |
/api/memories/search |
POST | Semantic search |
/api/memories/:id |
GET | Get a single memory |
/api/memories/:id |
DELETE | Delete a memory |
/api/consolidate |
POST | Run deduplication and cleanup |
/api/conflicts |
GET | Get detected memory conflicts |
Web Dashboard
Engram includes a built-in web dashboard at http://localhost:3838 when running the full server:
- Dashboard — Overview of memory stats and recent activity
- Memory Browser — Browse, filter, and manage all stored memories
- Search — Semantic search with similarity scores
- Statistics — Charts and breakdowns by category, namespace, and time
- Agents — Integration hub with a setup wizard for connecting MCP clients
How It Works
-
Store: When an AI agent calls
engram_remember, the memory text is embedded locally using all-MiniLM-L6-v2 (a 23 MB model that runs on CPU). The embedding and metadata are stored in a local SQLite database at~/.engram/memory.db. -
Recall: When
engram_recallis called, the query is embedded with the same model and matched against stored memories using cosine similarity. FTS5 keyword matching runs in parallel, and results are merged using a hybrid scoring algorithm. -
Deduplicate: Before storing, Engram checks existing memories for similarity. Exact duplicates (>0.95) are rejected. Near-duplicates (0.92–0.95) are merged intelligently.
-
Learn: The
engram_feedbacktool lets agents mark memories as helpful or unhelpful. This adjusts confidence scores and influences future recall ranking. -
Protect: Every memory passes through secret detection before storage. API keys, passwords, tokens, and other sensitive data are automatically blocked.
Configuration
Engram stores its data and config in ~/.engram/:
~/.engram/
├── memory.db # SQLite database (memories + embeddings)
├── config.json # Server configuration
└── models/ # Cached embedding model (~23 MB)
Default settings work out of the box. To customize:
// ~/.engram/config.json
{
"port": 3838,
"defaultNamespace": "default",
"recallLimit": 5,
"confidenceThreshold": 0.3,
"secretDetection": true
}
Advanced Usage
Namespace Isolation
Organize memories by project or client:
# Store memories in different namespaces
engram remember "Uses Next.js 14 with app router" --namespace my-saas
engram remember "WordPress multisite with Redis cache" --namespace client-site
# Recall searches within a namespace
engram recall "what framework?" --namespace my-saas
AI agents can use namespaces automatically — just include the namespace parameter in engram_remember and engram_recall calls.
Temporal Queries
Filter memories by time:
engram recall "deployment changes" --after "last week"
engram recall "API decisions" --after "2025-01-01" --before "2025-06-01"
Export for Documentation
Export your project's memory as documentation:
engram export --format markdown --namespace my-project > PROJECT_CONTEXT.md
engram export --format json > memories-backup.json
Contributing
See CONTRIBUTING.md for development setup and guidelines.
git clone https://github.com/HBarefoot/engram.git
cd engram
npm install
npm run dev
License
MIT © 2026 HBarefoot
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.