Chat Seek v0.3.4

Chat Seek now has a collapsible Options section below the search box. Choose Auto, Find a chat, Exact terms in originals, Quick answer, or Complete answer scan. The selection is remembered across views and restarts.

Exact terms runs without Laya or an answer API. Quick answer stops after the early original-text passes. Complete answer scan keeps the full archive pass. Original-text excerpts sent to Laya are kept within its token budget.

The Linux x64 VSIX supports VS Code 1.99+ and VS Code Remote WSL.

Chat Seek v0.3.3

Question search now checks the complete original text of likely chats before the long Laya scan. When ripgrep is available, it also searches the full transcript archive for distinctive literal terms. This finds answers hidden beyond the 2,800-character normal-search index limit. The focus window now looks for query terms together across the full message.

PIN and passcode questions stay local. An unambiguous numeric code tied to the named account is copied from the transcript with an exact citation, even when Laya would reject that chunk. Other matching citations can continue to appear as the scan progresses. No PIN or passcode question sends chunks to an answer provider, and labeled PINs are redacted from optional summaries.

On the owner's local archive, the BenchmarkHeaven DM question found its exact source citation in roughly 1.5–2.8 seconds with Laya loaded. The PIN itself was not included in logs or this release.

Chat Seek v0.3.2

Question search now shows Laya-matched chunks immediately while up to two answer requests run in parallel with continued scanning. Verified answers move to the top with their exact source quote; candidates rejected by extraction disappear. A bounded queue keeps requests under control, Stop clears pending candidates, and the search resumes visibly when you reopen the view. Question searches reuse the current index instead of rebuilding it each time.

Install the Linux x64 VSIX in the VS Code environment containing your chat histories. Existing users can reload their window after updating. Search and Laya stay local; optional answer extraction still requires a configured provider and explicit consent.

Chat Seek 0.3.1 — faster question search

Faster question-search startup

Question search now checks likely original transcript messages first. This quick pass uses the normal index only to find promising locations; the text sent to Laya is read from the original files. The exhaustive pass then streams each chat backward in overlapping chunks, so it starts reporting checked chunks without preparing an entire large transcript. Progress distinguishes the quick pass from the full archive scan and updates while reading large source files.

The full scan still checks every supported original transcript chunk and can take many hours on large archives. Use Stop after finding the answer if you do not need the rest of the scan.

Explainer

The narrated 60-second video explains the private index, normal keyword shortlist and Laya reranking, and question search. It uses illustrative data.

Linux x64 VSIX for VS Code 1.99+ and Remote WSL is attached.

Chat Seek 0.3.0 — cited question search

Question search

Chat Seek now recognizes factual questions with local Laya and scans original Claude Code, Codex, and OpenCode transcripts in overlapping, Laya-sized chunks. It visits newest chats and newest chunks first, streams results, shows progress, and supports Stop.

With separate opt-in consent, GPT-5.6-Luna or a configured fallback provider extracts a short answer from a matching chunk. Every displayed citation must be an exact substring of that chunk. You can expand the full chunk, open it in VS Code, read nearby chat context, or resume the session in its CLI.

Description search and optional cached chat summaries remain available. Question scanning is exhaustive and may take a long time on large histories.

Linux x64 VSIX for VS Code 1.99+ and Remote WSL is attached. Other platforms can build from source with the matching ONNX runtime.

Chat Seek 0.2.1 — GPT-5.6 Luna summaries

Use GPT-5.6 Luna by default for OpenAI and OpenRouter chat summaries. Live synthetic summary verified with OpenAI. Summaries remain opt-in.

Chat Seek 0.2.0 — summaries, dates and resume

Chat Seek 0.2.0

• Optional cached one-sentence chat summaries, with OpenAI, OpenRouter, TensorX and custom OpenAI-compatible providers.
• Summaries are opt-in; API keys use VS Code SecretStorage or environment variables.
• Human-readable last-activity dates with exact dates on hover.
• Resume Claude Code, Codex and OpenCode sessions in VS Code integrated terminals.
• Activity Bar search sidebar, expandable excerpts and a pin-search-tab action.
• Agent installation instructions and local-environment adaptation guidance.

Verified: 10 automated tests, syntax checks, responsive webview checks at 360px and 1000px, synthetic live summary calls to OpenAI, OpenRouter and TensorX, and installation in VS Code Insiders on WSL. CLI resume arguments were verified; interactive session resumption was not exercised.

The attached VSIX targets Linux x64, including WSL, and requires VS Code 1.99+. Summaries remain disabled until enabled explicitly.

Chat Seek 0.1.2

Faster local Laya reranking, automatic index refresh, and improved OpenCode conversation navigation. Linux x64 / VS Code Remote WSL build.

Chat Seek 0.1.1

Refreshes the local chat index when search opens and fixes navigation for OpenCode results. Includes local Laya reranking. Linux x64 / VS Code Remote WSL build.

Chat Seek 0.1.0

Search local Claude Code, Codex, and OpenCode chats in VS Code. Linux x64 / VS Code Remote WSL build; includes local Laya reranking. The first Laya use downloads model weights separately.