AI Automation Suggester 1.2.1

AI Automation Suggester 1.2.1

Release Date: 2025-01-05

Highlights

• Improved Prompt Management & Token Handling

  • Added approximate token counting and truncation for both OpenAI and Google calls to avoid sending overly large prompts.
  • Ensures requests stay within each provider’s maximum token limit (e.g., 30,720 tokens for Google, 32,768 for OpenAI).

• Model-Specific Parameter Adjustments

  • For OpenAI:
    • Standard models continue to use max_tokens.
    • New or special models (e.g., gpt-4o, o1, o1-mini, o1-preview) that require max_completion_tokens are now properly supported.
    • Prevents the “Unsupported parameter: 'max_tokens'” error on models that do not accept max_tokens.
  • For Google:
    • Simplified approach that uses maxOutputTokens only—removed checks for non-Google models (e.g., gpt-4o, o1-preview).

Detailed Changes

1. OpenAI Fixes

   • Added logic to detect specific models needing max_completion_tokens instead of max_tokens (e.g., gpt-4o, o1, o1-mini, o1-preview).
   • When prompt size is too large, we truncate it to stay within a safe limit (default ~32K tokens).

2. Google API Updates

   • Introduced token counting to cap request size at 30,720 tokens, addressing the INVALID_ARGUMENT errors.
   • Removed references to non-Google models in the Google request code.

3. General Resilience

   • Safer request-building steps guard against sending huge prompts or invalid parameters.
   • Logging improvements to warn when prompts exceed size limits.

Upgrading

• No special steps required for upgrading from 1.2.0 to 1.2.1.
• Restart Home Assistant after updating to ensure the changes take effect.

Notes

• If you continue to see truncation warnings, consider refining your prompts or summarizing content.
• If you use custom models not listed here (beyond gpt-4o, o1, o1-mini, o1-preview), you can extend the logic in coordinator.py similarly to handle model-specific parameters.

---

Thank you for using AI Automation Suggester!