Skip to main content

Changelog

All notable changes to CommandLane are documented in this file.

v0.2.1 (2026-03-21)​

Fixes​

  • Auto-scroll to config section when clicking Connect on integrations
  • Download button navigating away instead of downloading
  • Notion links in Captures page now open correctly via openUrl
  • Removed VOICE.md requirement from agent startup
  • Watchdog import fallback for frozen (packaged) builds

Improvements​

  • Website downloads now served from GitHub releases
  • Download analytics tracking via Vercel Analytics
  • Website switched from waitlist to direct download

v0.2.0 (2026-03-20)​

First stable release built on the new Tauri + React desktop architecture. Major feature additions including the Banner system, deep agent upgrade, MCP integration, and cognitive memory.

  • Background tasks - Schedule recurring AI tasks with real-time feed, approval workflows, and session-level capability grants
  • Approval flow - Agent requests permission before executing sensitive actions; you approve, defer, or dismiss
  • Reminders - Due reminders surface directly in the Banner feed
  • Banner Pro - Premium tier with 14-day free trial (free during beta)

AI Agent — Major Upgrade​

  • 20+ tools - Shell execute, file operations, directory listing, workspace file read/write, Notion sync
  • @mention subagents and /slash command routing
  • Conversation persistence for multi-turn memory across sessions
  • Sessions page - List and resume past agent conversations from the dashboard
  • A-RAG hierarchical retrieval tools for knowledge base search
  • Ollama support - Use local models via Ollama
  • Ghost-text autocomplete for @agents and /commands
  • Context compaction to stay within model context limits
  • Tool budget guardrails - Prevents wasting calls on failed/rate-limited tools
  • Credential scrubbing and awareness for safe tool execution

MCP Integration (NEW)​

  • Production MCP runtime with validation and flexible auth
  • Provider catalog - Brave Search, Google Workspace, Vercel, Atlassian
  • stdio MCP transport for Claude Code integration
  • MCP package (@commandlane/mcp) for Claude Desktop and Claude Code

Cognitive Memory​

  • Importance scoring, automatic extraction, tiered search, and time-based decay
  • OpenAI embeddings wired for semantic search
  • Context-aware storage with self-detection skip

Security Hardening​

  • Restricted process token sandbox for shell commands
  • Workspace integrity monitoring for agent sessions
  • Operation-level WebSocket authorization
  • Resume-time sanitization for conversation context isolation
  • Sensitive file read blocking in auto-allowed shell commands
  • Production-grade WebSocket connection resilience

Desktop & UI​

  • New desktop shell - Full migration from Tkinter to Tauri + React with system tray
  • Light theme - Off-white theme option with one-click switching
  • Sessions page for agent conversation history
  • Dynamic metrics from Notion database config
  • Setup wizard with folder picker
  • User data deletion option on uninstall

Payments & Licensing​

  • Stripe checkout flow with success/cancelled pages
  • Banner Pro license with 14-day free trial

Infrastructure​

  • Monorepo reorganization (_website, _docs-site)
  • Agent server crash fix for Nuitka production builds
  • Keyring credential cleanup on uninstall
  • Website SEO metadata across all routes
  • Download tracking via Vercel Analytics

v0.1.3​

Agent Security Audit​

  • Tiered Shell Permissions - Commands classified as auto-allow, require-approval, or blocked with !/!! prefix support
  • Rate Limiting - Per-turn and per-hour limits on dangerous tool invocations
  • Write-Then-Execute Prevention - Agent-written files tracked via manifest and blocked from execution
  • WebSocket Token Auth - Session token authentication for agent server connections
  • MCP HTTPS Enforcement - External MCP servers require HTTPS (localhost exempt)
  • Conversation Auto-Purge - Sensitive tool results purged from stored conversation history
  • Persistent Audit Log - All tool invocations logged to SQLite for accountability
  • Context Sanitization - XML tag escaping prevents prompt injection via agent context
  • Exact-Match Command Approvals - Approved commands use full-string match with 24h TTL expiry
  • MCP Tool Collision Detection - Built-in tools take precedence over MCP tools with same name

Earlier Releases​

Performance & Reliability​

  • Background Worker Integration - Fixed queue reliability with retry logic (3 attempts, 50ms delay)
  • Database Corruption Fix - Disabled FTS5 UPDATE trigger to prevent corruption on rapid captures
  • Classification Speed - TinyBERT averaging 13.3ms with 99% accuracy

Architecture​

  • Hybrid Search - RRF fusion of FTS5 + semantic search (<500ms @ 10k entries)
  • Three-Tier Classification - Cache → Rules → TinyBERT/LLM hierarchy
  • SQLite Hybrid Storage - FTS5 + sqlite-vec for scalable local storage

Roadmap​

Upcoming features:

  • Mobile Companion - iOS/Android quick capture apps
  • Plugin System - Extensible architecture for custom integrations
  • Smart Collections - AI-powered automatic grouping of related entries
  • MCP Server Marketplace - Discover and install community MCP integrations

Contributing​

Found a bug or have a feature request?