Chat Searcher

Never lose an AI conversation again. Chat Searcher indexes every chat across all your workspaces and lets you find any past conversation in seconds.

Works with Cursor, Windsurf, and other VS Code-based IDEs.

Why?

AI coding assistants don't have a built-in way to search your chat history. Once a conversation scrolls away, it's gone. Chat Searcher fixes that -- it reads your IDE's local storage, indexes everything, and gives you instant full-text search across all your projects.

Features

• Instant full-text search -- find any conversation by keyword using the BM25 ranking algorithm
• Lightning-fast indexing -- indexes hundreds of conversations in ~4 seconds using a single optimized SQLite query
• Cross-workspace -- searches across all your projects at once, with filtering by workspace
• Full conversation view -- click any result to see the complete chat with all messages
• Correct roles -- properly distinguishes user messages from assistant responses
• Copy to Chat -- bring any past conversation into your current workspace's chat
• Sort by relevance or date -- find the most relevant or most recent conversations
• 100% local -- all data stays on your machine. Nothing is sent anywhere.
• Low memory footprint -- uses the native sqlite3 CLI to query large databases without loading them into memory

Quick Start

1. Install the extension
2. Press Cmd+Shift+H (macOS) or Ctrl+Shift+H (Windows/Linux)
3. Type your search query
4. Click a result to view the full conversation

That's it. The index is built automatically the first time you open the search panel.

Commands

Command	Shortcut	Description
Chat Searcher: Search Chats	Cmd+Shift+H / Ctrl+Shift+H	Open the search panel
Chat Searcher: Reindex All Chats	--	Rebuild the index (run after new conversations)

How It Works

1. Reads composer metadata from each workspace's local state.vscdb (small SQLite files, ~100KB each)
2. Fetches all conversation bubbles from the global state.vscdb in a single query using json_extract -- extracting only the fields needed (role, text, timestamp), not the full multi-KB JSON blobs
3. Groups bubbles by conversation and builds a BM25 search index in memory
4. Search queries return results ranked by relevance with highlighted snippets

The global storage database can be 2+ GB, but Chat Searcher never loads it into memory. It uses the system sqlite3 CLI for zero-copy querying, with a fallback to sql.js for systems where the CLI isn't available.

Supported IDEs

• Cursor
• Windsurf
• Any VS Code-based IDE that stores AI chat in workspaceStorage/globalStorage

Requirements

• macOS, Linux, or Windows
• sqlite3 CLI (pre-installed on macOS and most Linux distros; optional on Windows -- falls back to in-memory loading)

Privacy

Chat Searcher is fully offline. It reads only local SQLite files on your machine. No network requests, no telemetry, no data collection.

License

MIT -- Alexander Goldberg