Engram provides persistent memory for AI coding agents using a single Go binary backed by SQLite with FTS5. It saves, searches, and synchronizes agent context across sessions via MCP, CLI, HTTP API, and TUI interfaces.
Project overview
It ships as a zero-dependency Go binary that stores agent memory locally in SQLite with full-text search and exposes twenty memory management tools to any MCP-compatible coding agent.
Project type
MCP · RAG · Infrastructure
Use cases
Knowledge Q&A · Coding & Development
Deployment
Refer to project documentation
License
MIT
Best for
Developers using MCP-compatible coding agents who need to persist and search session context locally.
AI engineers looking to export agent memory to an Obsidian vault knowledge graph.
Key capabilities
Exposes 20 memory management tools via MCP over stdio transport for any MCP-compatible agent.
Persists agent memories to a local SQLite database with FTS5 full-text search indexing.
Provides a local HTTP REST API for session tracking and memory management.
Command-line interface for saving, searching, exporting, and managing memories.
Interactive terminal dashboard for browsing and searching memories.
Shares memories across machines using compressed chunks without merge conflicts.
Replicates local memories to a shared cloud instance for team access.
Detects conflicting architectural memories and uses LLM reasoning for semantic similarity checks.
Limitations and risks
MCP transport is stdio only, with no HTTP or network MCP endpoint available.
The HTTP API binds to 127.0.0.1 only and cannot be reached from inside a container out of the box.
Semantic scan does not discover totally lexically unrelated pairs on its own.
Known repairable cloud sync, upsert, or canonicalization failures may occur, requiring an explicit doctor or repair flow.
Getting started
Installs via a single Homebrew command followed by an automated agent setup command.
After installation, running the agent setup command writes the required MCP config and plugin files automatically.
Evidence and sources
README: A **Go binary** with SQLite + FTS5 full-text search, exposed via CLI, HTTP API, MCP server, and an interactive TUI. Works with **any agent** that supports MCP
README: What `engram setup` does — it writes MCP config and plugin files for the chosen agent. After setup, restart your agent and it is ready. No server to start manually.
README: `engram serve` is only needed when a plugin uses the HTTP API for session tracking
README: Full tool reference with parameters → [DOCS.md#mcp-tools-20-tools](DOCS.md#mcp-tools-20-tools)
README: Agent calls mem_save → title, type, What/Why/Where/Learned 3. Engram persists to SQLite with FTS5 indexing 4. Next session: agent searches memory, gets relevant context