v1.0.318

Date released: August 13, 2026

Enhancements

AI

• Document Processing Jobs Now Run via Queue: Parsing, classification, and extraction workflows for documents are now processed through a managed jobs queue with heartbeat monitoring and deterministic workflow IDs. Stuck jobs are detected and terminated within minutes rather than running indefinitely, and duplicate processing on retry is prevented automatically.

Bug Fixes

• Fixed an issue where agents with many MCP tools connected could silently lose access to workbook write operations — such as adding or updating rows — due to tool selection pruning. Write tools are now always available, and for agents connected to many workbooks, the available write capacity is distributed fairly rather than front-loaded.
• Fixed a high-priority issue where creating a page-level filter on a dashboard resulted in a blank filter name and the filter not appearing on the dashboard at all. Filters now save and display correctly after configuration.
• Fixed an issue where the Python Agent failed on ETL workbooks (MV + Last Mile) because it was still pointing to the old S3 storage location after the GCP migration. The agent now reads Parquet files from GCS correctly using per-company storage keys.
• Fixed a race condition where completing a warehouse pipeline could leave a brief window during which concurrent pipelines reading from the same BigQuery table would fail with a "table not found" error. The external table is now recreated immediately after being dropped, closing that gap.