🪶 Paradigm Version: Quiet-Quail 0725

The Quiet Quail update brings enhanced privacy controls, smarter document handling, and new integration capabilities. These improvements aim to support more autonomous, secure, and flexible usage across a wide range of workflows.

1. Document Handling Enhancements

🌐 Web Scraping Datasource (Beta)

Users can now ingest website content as a structured knowledge source.
Core features include:

  • Intelligent crawling and markdown extraction.

  • Configurable by users via URL input and webpage specific parameters such as .
  • Detect and retrieve linked files of supported filetypes.
  • All extracted content is ingested for availability accross tools in the platform.

📄 HTML, XHTML, TXT Format Support

Uploading .html, .xhtml, .txt documents is now fully supported. This resolves previous compatibility gaps, especially for legacy or low-footprint document pipelines.

  • Error handling for unsupported formats has been improved.
  • Datasources automatically syncs new format type

🧠 Smart Parsing 

Document ingestion now automatically selects the optimal parser based on file type. This provides better extraction results while reducing manual intervention.

  • .pdf, .docx default to legacy parser with fallback logic.

  • .html, .pptx, .png, .jpg, .jpeg use specialized vision-based parsers.

  • .md, .csv, .xlsx routed through dedicated text pipelines.

  • Behavior is configurable for benchmarking or debugging.


2. Privacy and Confidential Sessions

🔒 Ephemeral Mode — Full Session Privacy

A new “Ephemeral Mode” allows users to start private sessions where no user inputs, document links, or feedback are stored. This is ideal for sensitive or exploratory conversations.

  • Ephemeral Mode can be toggled at company level by system admins.

  • When enabled: Sessions & Documents are not persisted or visible to admins. 

  • Sessions are automatically deleted after 15 minutes of inactivity.


3. Additional Improvements

  • Bug reporting UX has been improved:

    • Comments are now mandatory.

    • Users can optionally preview and share associated documents.

    • Flags are hidden when in Private Mode.

  • Chat sessions can now visually indicate long-running background tasks in the UI.

  • A disclaimer ("Alfred may make mistakes") is now permanently shown below the chat input for transparency.


4. Alfred 4.2: Visual Language Model (VLM)

🧠 Alfred Goes Multimodal
Version 4.2 of Alfred introduces a major upgrade: it is now a Visual Language Model (VLM). This means it caninterpret visual elements within documents—significantly improving its understanding of complex, hybrid formats.

Key Highlights:

  • 🖼️ Smart interpretation of embedded images: photos, diagrams, charts.

  • 📊 Improved analysis of tables and visually structured data.

  • 🧭 Enhanced contextual comprehension for documents mixing text and visuals.

  • 🛠️ Fully integrated into the automatic parser selection system (see Section 1.3).