Chat Seek

What it does

• Plain-language search across Claude Code, Codex and OpenCode session files — user and assistant text only, indexed locally
• Laya local decision-model reranking on top of keyword results, with a switch (chatSeek.useLaya) to fall back to pure keyword search
• Resume in one click: the matching CLI reopens in VS Code's terminal with its session ID and original working directory when available
• Optional AI summaries are off by default and defensive: sampled excerpts only, credential redaction, per-provider disclosure, SecretStorage keys, cached locally by chat content

When to use it

Developers whose valuable Claude Code sessions vanish into ~/.claude/projects and who want to retrieve, reread and resume them without leaving the editor — especially on Linux/WSL where the prebuilt package installs directly.

Getting started

# 1. download chat-seek-linux-x64-0.2.1.vsix from the latest release
# 2. in VS Code (1.99+): Extensions: Install from VSIX...

# or build on Linux x64 (Node.js 20+):
git clone https://github.com/fstandhartinger/chat-seek-vscode.git
cd chat-seek-vscode && npm ci && npm test && npm run package
code --install-extension chat-seek-linux-x64-0.2.1.vsix --force

# open it: Cmd/Ctrl+Shift+P -> "Chat Seek: Search past AI chats"

Things to know

The prebuilt VSIX is Linux x64 only (including VS Code Remote WSL) — other platforms must change the vsce target and rebuild. First Laya use downloads about 1.7 GB of model weights into ~/.cache/receptron-laya and runs CPU inference; disable it via chatSeek.useLaya on weak machines. Enabled summaries send sampled excerpts to your configured provider and may incur charges. Only user and assistant text is indexed — tool results and subagent transcripts are excluded — long messages are clipped at 2,800 characters, and OpenCode's newer database-only storage is not yet imported. Keyword shortlisting depends on word overlap, so chats sharing no vocabulary with your description can be missed. MIT.