AI Chat Storage & Workspace Manager

Manage local AI chat history and related developer tooling from one VS Code dashboard. Inspect, search, preview, export, and clean chat sessions across supported AI tools, audit project instructions, review MCP server configuration, track storage trends, and build a local Markdown knowledge base.

Copilot chat history lives in workspaceStorage/<hash>/chatSessions/*.jsonl and is never cleaned up automatically. Old sessions from abandoned branches, renamed folders, or projects you deleted months ago just... sit there, sometimes adding up to hundreds of megabytes and slowing VS Code's startup and search indexing.

Features

• One dashboard, every workspace — scans Code, Code - Insiders, and VSCodium storage and groups sessions by project, biggest first.
• Overview and next steps — summarizes storage, security scan status, project instruction gaps, MCP inventory, and analytics readiness with direct links to each workflow.
• Read-only demo mode — previews synthetic data across every tab, including Rules and MCP, without requiring project files or server configuration. Demo actions never scan, edit, export, delete, or contact endpoints.
• Real chat titles — recovers the actual renamed title (or a preview of the first message) instead of a bare UUID.
• Transcript preview — inspect a bounded, escaped transcript in the dashboard before exporting or deleting it.
• Global search — find a chat by title or filename across all projects at once, with inline bulk delete.
• Orphaned workspace detection — flags chats belonging to folders that no longer exist on disk, with a one-click cleanup.
• Quick Clean — delete everything older than 30 / 60 / 90 days in one action.
• Full-transcript export — before deleting, export selected chats as readable Markdown files (one per chat) or a single structured JSON file, with actual message content, not just metadata.
• Optional auto-backup — turn on aiChatCleaner.autoBackupBeforeDelete and every delete is preceded by an automatic transcript backup, as a safety net.
• Security review — scan supported chat history for likely credentials, inspect matches with values redacted, open the source, remove an unwanted chat, or export a redacted report. Placeholder values in examples are filtered.
• Project-aware AI Rules Manager — inspect common root manifests and README guidance, identify missing instruction areas, and create a stack-aware starter without overwriting existing rules. Prose sync excludes structured Continue config files.
• MCP inventory and HTTP checks — review configured servers across supported tools. HTTP reachability checks are manual and time-limited.
• Storage analytics — track daily snapshots, overall growth, change since the previous snapshot, and the tool with the largest increase.
• Repeatable knowledge-base builds — regenerate marked chat notes in place to avoid duplicate copies while preserving user-authored indexes and files.
• Theme-native UI — built entirely from VS Code's own CSS variables, so it matches your light/dark/high-contrast theme instead of looking like a bolted-on webpage.
• Sort by newest, oldest, largest, smallest, or title; per-project or global bulk actions; confirmation dialogs you can turn off in settings.

Usage

1. Run AI Chat Cleaner: Open Dashboard from the Command Palette.
2. Start from Overview for recommended next steps, or click Preview demo to explore synthetic examples across all tabs. Click Exit demo to return to your data.
3. In Storage, pick a project or search everywhere, preview transcripts, and select chats to export or delete.
4. Use Security to review redacted findings, AI Rules to audit the opened project, MCP Servers to inspect config, Analytics to review trends, or Knowledge to build a local vault.

Command Palette also has AI Chat Cleaner: Export All Chat Backups, which backs up every chat session found on the machine without opening the dashboard.

Settings

A note on the storage format

Copilot's on-disk chat session format (kind: 0/1/2 transaction log entries) isn't officially documented and has changed between VS Code releases. This extension parses the shapes seen in current VS Code builds and falls back to a generic text-recovery pass for anything it doesn't recognize, so an export should never come back empty — but if you're on a very new or very old VS Code build and something looks off, please open an issue with your VS Code version.

For maintainers

The Marketplace listing's Overview is generated from this README. The dashboard's in-app Overview is different: it summarizes the current machine's local data and suggested actions.

The published extension identifier is OrangeAI.vscode-copilot-chat-cleaner. Its publisher ID is already set in package.json; do not replace it with a placeholder.

To verify and package a local build:

npm test
npx @vscode/vsce package

For a release, increment the version in package.json and package-lock.json, run the checks above, then upload the generated VSIX from the Marketplace publisher management page. Alternatively, authenticate for the OrangeAI publisher and run npx @vscode/vsce publish. Users install the published extension directly from the Marketplace; they do not need to package it themselves.

Privacy

Scanning, transcript preview, instruction analysis, analytics, and knowledge-base generation run locally. The extension does not send this data to an AI service. If you manually click Test for an HTTP MCP server, the extension sends a short-timeout GET request to that server's configured URL. Exports and knowledge-base files are written only to locations you choose; deleting chats may first create a local backup when enabled.