← Back to changelog

⚙️ Config v1.3.5

v1.3.5

  • Added document_parser OCR strategy for local text extraction

    • Extracts text from PDF, DOCX, XLS/XLSX, and OpenDocument files without external OCR services
    • Uses pdfjs-dist, mammoth, and SheetJS locally — no API key needed
    • Runs automatically for agent file uploads even without any ocr configuration, and as a fallback when a configured OCR strategy fails
    • Can also be set explicitly via ocr.strategy: "document_parser" in your librechat.yaml
    • See OCR Configuration for details

  • Added thinkingLevel parameter for Gemini 3+ models

    • Controls the thinking effort level: "minimal", "low", "medium", or "high"
    • Exposed as a slider in the Google endpoint settings UI
    • Applies to Gemini 3+ models only (Gemini 2.5 models continue to use thinkingBudget)

  • Added reasoning_effort parameter for Bedrock models

    • Configures reasoning effort for supported Bedrock models (ZAI, MoonshotAI)
    • Options: "low", "medium", "high"
    • Exposed as a slider in Bedrock endpoint settings UI
    • See AWS Bedrock Configuration for details

  • Added Bedrock document upload support

    • Users can now upload documents (PDF, DOCX, etc.) when using Bedrock endpoints

  • Updated default model lists

    • Added gemini-3.1-pro-preview, gemini-3.1-pro-preview-customtools, gemini-3-pro-preview, and gemini-3-flash-preview to Google default models

  • Changed date template variable format

    • {{current_date}} now uses named weekday (e.g., 2026-03-04 (Monday)) instead of numeric index
    • {{current_datetime}} now includes timezone offset (e.g., 2026-03-04 14:30:00 -05:00 (Monday))

  • Updated OpenRouter reasoning configuration

    • Reasoning parameters are now sent as a unified reasoning object to match OpenRouter's updated API