cass (coding-agent-search) — A Searchable Memory for Your AI Coding Sessions

What it is: A high-performance Rust tool that aggregates the saved conversation history of 30+ different AI coding agents (Claude Code, Codex, Cursor, ChatGPT, Copilot, and many more) into a single, unified, searchable timeline. It gives you a terminal interface (TUI) or a JSON-based CLI to find past debugging sessions, solutions, and context in milliseconds.

Why it's interesting

This project solves a really common pain point. Many developers use several AI coding assistants, and each one stores its history in its own proprietary format (JSONL, SQLite, markdown). When you solve a tricky problem in Cursor, that knowledge is locked away and invisible to you when you switch to Claude Code. cass treats your combined agent history as a single, persistent knowledge base.

It's also a serious, well-engineered tool with an unusually forward-looking feature: Robot Mode. Instead of just being a search tool for humans, it exposes a stable JSON API specifically designed for other AI agents to call. This lets your current coding agent autonomously search the historical sessions of all your past agents — effectively giving it access to a shared memory of previous work, which is a very practical step toward agent persistence and continuity.

Interesting technical details: Written in Rust, it uses edge N-gram indexing for sub-60ms, search-as-you-type full-text queries. It supports optional local semantic search (via a small MiniLM embedding model) with no network calls at query time and automatic fallback to lexical search if the model isn't installed. It's designed defensively — atomic index swaps so readers never see a torn index, quarantine of corrupt assets rather than deleting them, and a doctor/diag surface for inspecting and repairing derived indexes.

Why it matters

It turns ephemeral AI output into a durable institutional asset. Instead of losing the brilliant debugging session from two weeks ago, you can find it. And because it exposes a machine-readable interface, it allows for a new class of workflows — like letting your current AI agent 'learn from' the accumulated problem-solving history of all your previous agents.