Skip to content

v0.26.2

Latest

Choose a tag to compare

@github-actions github-actions released this 29 Jun 01:59
· 38 commits to main since this release

Summary

Released on June 29, 2026.

New features

  • Chat channels:
    • Integrates WhatsApp via a QR code scan. #16238
    • Integrates DingTalk via bot API credentials. #16183
    • Integrates WeCom via WebSocket connection. #16175
  • File parsing: Adds fallback logic for PP-OCRv6 and similar text recognition models, and integrates image parsing within the PaddleOCR pipeline. #16150

Improvements

  • Datasets
    • Adds an edge-case check to the file ingestion logic to safely link empty folders to a dataset. #16296
    • Adds pagination to the get_flatted_meta_by_kbs() method to prevent CRUD failures relating to datasets with more than 10,000 documents. #16095
  • Chat channels: Ensures end-user conversation histories persist across restarts but still separate when a channel is bound to a new dialog. #16274

i18n

  • Expands i18n coverage and refines translations for the Korean UI. #16203
  • Adds missing French translations for chat channels, username validation, and model editing. #16217

Bug fixes

  • File parsing:
    • A ValueError was raised during .docx parsing on 'Heading' styles. #16284
    • Tables in .docx files were silently removed by the laws document parser. #16155
    • Lone Markdown headers were separated from their subsequent sections when a DeepDoc delimiter was set. #16109
    • Markdown tables appeared twice in chunk results, once as a standalone chunk and again within a regular text chunk. #16143
  • MCP
    • The MCP server hung indefinitely when encountering empty document pages; the final page of documents was silently dropped due to flawed pagination. #16285
    • The MCP connection and dataset discovery process failed because the server's hardcoded fetch limit exceeded the introduced page_size restriction. #16148
  • Dataflow
    • Restores the dataflow rerun endpoint and ensures the ingestion response includes the DSL payload to reliably render the timeline and parser views. #16292
    • Restores dataflow parser defaults and returns the actual SSE payload, ensuring pipeline runs correctly surface message IDs and log updates. #16290
  • LLM: Failed to use the new gemini-3.5-flash model via the Google Cloud (Vertex AI) provider using eu or us region endpoint. This is resolved by explicitly multi-region routing requests to the required aiplatform.<region>.rep.googleapis.com domains instead of synthesized <region>-aiplatform.googleapis.com hosts. #15990
  • UI/UX: The metadata add modal sent empty values to the backend. #15229

What's Changed

New Contributors

Full Changelog: v0.26.1...v0.26.2