Natively-AI-assistant/natively-cluely-ai-assistant v2.3.0

Natively 2.3.0

on GitHub

Summary

Major update introducing Automatic Language Detection for Speech-to-Text, a beautifully revamped automated email ecosystem, and massive production hardening for the Natively API.

What's New

• Auto-Language Detection: Added a new "Auto" language option. The system now seamlessly identifies the spoken language on-the-fly and intelligently routes the transcription to matching BCP-47 dictionaries. Works fully with Natively API STT, Deepgram, Google STT, ElevenLabs, and Whisper workflows.
• Automated Customer Emails: Built a complete transaction and marketing email engine to handle payment recovery, receipts, and API/Pro cross-sells automatically.

Improvements

• Vivid Editorial Email Design: Remastered all email templates featuring multi-layered glassy mesh-gradients, frosted CTA buttons, and WCAG AAA accessible high-contrast typography.
• Persistent Overlay Bounds: The interface now accurately remembers the overlay window position across hide/show cycles and gracefully handles multi-monitor coordinate math. (Thanks to Nakul Magotra - PR #158)
• Optimized Vision Analysis payloads: The LLM Helper now dynamically compresses screenshots via sharp before API transmission, ensuring multiple images can be analyzed instantly without hitting 413 Payload Too Large limits.
• API Scale Capabilities: Raised standard body payload limits and switched all application configurations to route through the primary canonical api.natively.software domain for maximal uptime.

Fixes

• Billing Engine Integrity: Fixed a race condition that could cause double-billing on Deepgram failover reconnects, and patched a vulnerability prohibiting search-session manipulation across API keys.
• Production Hardening (v2.4.0 & v2.4.1): Eradicated 11+ critical bugs ranging from ungraceful Railway restarts, Gemini 429 rate-limit thrashing, Groq cooldown memory leaks, to Webhook TOCTOU idempotency clashes.
• Security & Schema: Strengthened input validation to actively block HTML/SQL injection chars before reaching the AI system prompt. Closed timing-attack side channels on admin endpoints and cleaned up dead DB columns.
• Auto-Mode Corrections: Updated ElevenLabs to the correct _realtime WebSocket models and fixed an overlapping bug where auto-detected languages incorrectly bled into subsequent meetings.

Technical

• Autonomous Load Testing: Shipped a comprehensive stress-testing suite simulating 30+ concurrent WebSocket connections and forced failure scenarios, certifying failover waterfalls under extreme duress.
• High-Performance Memory: Replaced O(N) Webhook logs with optimized pre-allocated circular buffers, clamping parameters to prevent memory exhaustion attacks.
• Graceful WebSocket Hand-offs: Server updates now correctly transmit 1001 "going away" frames to all connected apps before shutting down, ensuring graceful reconnections during rolling deployments.

⚠️macOS Installation (Unsigned Build)

Download the correct architecture .zip or .dmg file for your device (Apple Silicon or Intel).

If you see "App is damaged":

• For .zip downloads:

  1. Move the app to your Applications folder.
  2. Open Terminal and run: xattr -cr /Applications/Natively.app

• For .dmg downloads:

  1. Open Terminal and run:

     xattr -cr ~/Downloads/Natively-2.3.0-arm64.dmg
     # Or for Intel Macs:
     xattr -cr ~/Downloads/Natively-2.3.0-x64.dmg

  2. Install the natively.dmg
  3. Open Terminal and run: xattr -cr /Applications/Natively.app

⚠️Windows Installation (Unsigned Build)

When running the installer on Windows, you might see a "Windows protected your PC" warning from Microsoft Defender SmartScreen saying it prevented an unrecognized app from starting.

Since this is an unsigned build, this is expected. You can safely ignore it by clicking More info and then Run anyway.