PhariaAI v1.260400.1

Added a 1-minute timeout for file upload polling in the Translation feature; if the timeout is reached, the user is offered an option to restart the process.

UI: Fixed an issue where a blank/white screen was shown when the application API was temporarily unavailable (e.g. during pod restarts); users are now shown an informative error message instead.

UI: Fixed an issue where users with an expired or invalid session token were shown a blank page; the token is now cleared automatically to allow a fresh login without getting stuck in a loop.

UI: Fixed an issue with incorrect font rendering in the Assistant.

UI: Fixed an issue where re-uploading a previously failed file in Translation did not trigger a new processing run; the previous error was shown instead, leaving users with no clear way to retry.

UI: Fixed an unclear error message shown when uploading a document with the same name but different content in the Translation feature.

UI: Improved the error message shown when uploading a scanned PDF that exceeds the OCR page limit; the message now clearly explains the issue in plain language.

Content: Updated the license information in the About section to correctly reference Llama 3.3 instead of Llama 3.1.

Backend: Fixed the Translation feature returning errors for non-admin users by adding the required ExecuteJobs permission to the deployment configuration.

Backend: Fixed an S3 permission issue causing document upload failures in restricted environments.

Backend: Updated retention periods for user upload stages in the Translation feature to be applied retroactively, bringing them in line with the intended 30-day policy.

Hide search_files tool if no included document is truncated

Convert strings that represent list of strings when Document Index MCP server is used

Prevent documents from other chats from being referenced accidentally

Uploaded documents are no longer forgotten when regenerating an answer

Run health sweep: A background worker now periodically checks for runs that have been stuck in a "running" state too long and automatically marks them as failed. This prevents zombie runs from silently blocking pipelines and gives users a clear terminal state instead of indefinite hangs.

Smart PDF parsing with OCR detection: The document pipeline now automatically detects whether a PDF page is digital text or a scanned image and routes each page to the right processing path. Mixed documents (e.g. a PDF with some native text pages and some scanned pages) are handled end-to-end without manual OCR configuration.

Re-trigger on previously failed dedup files: If an earlier transformation run failed on a deduplicated file, re-uploading or retrying it now correctly triggers a fresh transformation run. Previously the dedup logic could silently skip the retry.

Observability improvements: OpenTelemetry tracing was added to the Qdrant collection initializer, and Prometheus metrics were expanded with an updated Grafana dashboard, giving better visibility into pipeline health and latency.

OCR jobs stuck in "running": Workflows that include OCR could remain in a running state indefinitely due to heartbeat and threading issues in the Temporal workers. Parallel page handling with proper heartbeats and main-thread signal fixes ensure OCR pipelines now complete or fail cleanly.

Storage client permissions on internal vs. external buckets: Some environments hit permission errors when pipeline steps tried to read uploaded documents from object storage. The storage client now correctly distinguishes internal and external bucket configurations so S3 access no longer blocks ingestion.

DOC/PPT conversion failures: Uploading .doc and .ppt files could result in silent conversion errors that left runs in a bad state. Error handling in the parser helpers is now explicit, surfacing clear failures instead of swallowing them.

user accounts with new user id can be created even with pre-existing accounts with same email address

file_ids filter is applied correctly again

Reduce chance of agents passing wrong input format to file_ids parameter of search tool