Engram — Persistent Memory System for AI Agents

How It Works

Remember Decisions

When your agent makes a decision, store it:

engram remember "We chose PostgreSQL over MySQL for this project because of JSONB support"
engram remember "API responses always use snake_case, frontend uses camelCase"
engram remember "Never use ORM for complex queries - write raw SQL instead"

Automatic Recall

When used as an MCP server, the agent automatically queries Engram for relevant context based on the current conversation:

User: "Set up the database schema"
Agent: [queries Engram] → recalls PostgreSQL decision, snake_case convention
Agent: "Based on our previous decisions, I'll use PostgreSQL with snake_case columns..."

Full-Text Search

SQLite FTS5 provides blazing-fast full-text search across all memories:

engram search "database"  # finds all database-related memories
engram search "naming convention"  # finds style decisions

Memory Categories

Organize memories by type:

engram remember --category architecture "Microservices with event-driven communication"
engram remember --category style "Use Tailwind CSS, no custom CSS files"
engram remember --category decisions "Chose Stripe over PayPal for payments"

TUI Dashboard

Interactive terminal UI to browse, search, edit, and delete memories:

engram tui

Key Stats

• 2,300+ GitHub stars
• Single Go binary, zero dependencies
• SQLite + FTS5 full-text search
• 4 interfaces: MCP, HTTP API, CLI, TUI
• Works with any MCP-compatible agent

FAQ

Q: What is Engram? A: Engram is a persistent memory system that gives AI coding agents long-term memory across sessions, storing decisions, patterns, and context in a local SQLite database with full-text search.

Q: Is Engram free? A: Yes, fully open-source under MIT license. All data stays local on your machine.

Q: How is Engram different from CLAUDE.md files? A: CLAUDE.md is static and manually maintained. Engram is dynamic — agents can read and write memories during sessions, and full-text search surfaces relevant context automatically.