Engram — Persistent Memory System for AI Agents

What it is

Engram is a persistent memory system designed for AI coding agents. It stores memories in SQLite with full-text search and ships four interfaces: an MCP server, an HTTP API, a CLI, and a TUI (terminal UI). Any AI agent that supports MCP or HTTP can use Engram to remember information across sessions without external dependencies.

This tool targets developers who use AI coding agents daily and lose context between sessions. Engram gives any agent long-term memory, whether it is Claude Code, Cursor, Cline, or a custom agent pipeline. The token_estimate for this workflow is approximately 2,800 tokens.

How it saves time or tokens

Without persistent memory, developers re-explain project context, coding conventions, and past decisions at the start of every AI session. Engram lets the agent store and retrieve this information automatically, saving the tokens spent on repeated context-setting. The SQLite backend ensures fast retrieval without the overhead of running a vector database.

How to use

1. Install Engram via go install or download the binary
2. Add the MCP server configuration to your agent's .mcp.json
3. The agent can now store and retrieve memories across sessions

Example

{
  "mcpServers": {
    "engram": {
      "command": "engram",
      "args": ["mcp"]
    }
  }
}

# Install
go install github.com/GentlemanHu/engram@latest

# CLI usage
engram store 'Project uses Go-Zero for backend, Nuxt 3 for frontend'
engram search 'what framework does the project use'

# TUI mode
engram tui

Related on TokRepo

• AI memory tools — Compare persistent memory solutions for AI agents
• MCP integrations — Browse MCP server configurations for data tools

Common pitfalls

• Memory quality depends on what the agent stores; configure clear storage prompts to avoid noisy or low-value memories
• SQLite full-text search works well for keyword matching but lacks semantic similarity search for fuzzy queries
• The MCP server must be running for the agent to access memories; restart it if the agent reports connection errors