Engram

Engram gives AI coding agents a persistent memory that survives session boundaries. It is a single Go binary with no external dependencies (no Node.js, no Python, no Docker) that stores memories in a local SQLite database with FTS5 full-text search. Memories are exposed via an MCP stdio server, an HTTP API, a CLI, and an interactive terminal UI, making it compatible with any agent that supports MCP — including Claude Code, Cursor, Windsurf, VS Code Copilot, Gemini CLI, Codex, and OpenCode.

• Agent-agnostic MCP server — connect any MCP-compatible agent with a single one-liner setup command per agent.
• 18 MCP tools — covers save/update/delete, search/retrieve, session lifecycle, conflict surfacing, and utility operations.
• SQLite + FTS5 storage — all memories persist locally in ~/.engram/engram.db; no cloud account required.
• Interactive TUI — run engram tui for a Catppuccin Mocha-themed dashboard with vim-style navigation, search, and detail views.
• Git sync — export memories as compressed chunks and commit them to a git repo to share across machines with no merge conflicts.
• Opt-in cloud replication — Engram Cloud provides cross-machine sync and a dashboard; local SQLite always remains the source of truth.
• Conflict surfacing — mem_judge and mem_compare MCP tools, plus a --semantic LLM-judge scan, detect vocabulary-different but semantically conflicting memories that keyword search would miss.
• Obsidian export (beta) — run engram obsidian-export to generate an Obsidian knowledge graph from stored memories.
• Single binary install — install via Homebrew (brew install gentleman-programming/tap/engram) or download a pre-built binary for Windows, macOS, or Linux.
• Open source (MIT) — full source available on GitHub; contributions welcome via the documented workflow.