# traz

traz is a local-first developer memory layer that gives AI coding tools a shared brain. It establishes a highly searchable, persistent timeline stored locally and exposed via the Model Context Protocol (MCP), ensuring that subsequent AI sessions inherit the context of previous decisions.

## Key Features
- **Shared AI Memory**: Stops context loss between Claude Code, Cursor, and Aider.
- **Local-First SQLite**: Built in pure Rust. Zero cloud dependencies, zero latency, absolute privacy.
- **Native MCP Server**: Hook into your workflow with one command, delivering dense, compressed context.
- **Semantic Vector Search**: Hybrid search powered by local ONNX-accelerated MiniLM embeddings and FTS5.
- **Token Compressed**: Dense Output Formatting compresses context payloads by up to 75%.
- **Cuby (Context Pet)**: A virtual Tamagotchi pet built into the CLI that you can interact with and feed context.

## Core Usage
- **CLI Search**: `traz search "auth bug"`
- **Start MCP Server**: `traz mcp start`
- **Interactive TUI**: `traz tui`
- **Manual Event Logging**: `traz add "commit message"`

## Architecture
`traz` consists of several highly optimized crates:
- `traz-core`: Primary data models.
- `traz-db`: SQLite wrapper and ONNX execution via `fastembed-rs`.
- `traz-mcp`: Model Context Protocol implementation.
- `traz-tui`: Terminal UI via `ratatui`.

For more detailed technical documentation and API specifications, visit https://traz.mithilgirish.dev