engram v0.1.0
Persistent memory for AI coding agents. One binary. Local and private.
Highlights
- 7 knowledge layers: entities, sessions, documents, tool intelligence, session coordination, knowledge management, work management
- 21 MCP tools accessible from Claude Code, Cursor, Windsurf, or any MCP-compatible agent
- Single Rust binary with embedded SurrealDB — no external services required
- Local ONNX embeddings via fastembed — no API calls, total privacy
- Multi-session support with automatic daemon and conflict detection
- Full CLI for manual interaction with all layers
Quick Start
git clone https://github.com/ymeiri/engram.git
cd engram && cargo build --release
./target/release/engram init
./target/release/engram serveSee the README for MCP client configuration and the Setup Guide for Claude Code, Cursor, and Windsurf.
What's Included
| Layer | Capability |
|---|---|
| Entity Knowledge | Graph of repos, tools, services, concepts with relationships and observations |
| Session History | Decisions, events, rationale tracking across coding sessions |
| Document Search | Semantic similarity search over indexed markdown documents |
| Tool Intelligence | Usage pattern learning and success-rate-based recommendations |
| Session Coordination | Parallel session awareness and file/component conflict detection |
| Knowledge Management | Document registry with version detection and deduplication |
| Work Management | Projects, tasks, PRs with entity connections and scoped observations |
Requirements
- Rust 1.80+ (for building from source)
- macOS or Linux
Pre-built binaries coming in a future release.