Release Notes - 26.11.00
- AI Review Agent introduces AI-powered document classification directly into Nebula review workflows, using matter-specific review protocols to configure agents that classify documents at scale. Before an agent is published for production use, its coding is compared with independent human review through a blind validation workflow using precision, recall, richness, and confidence measures; versioning, human approval gates, citations, and audit logging provide transparency and defensibility throughout the process.
- AI Case Explorer Support:
- Agentic Chat can now find documents similar to a given document by its Document ID. Administrators can configure the number of results, the nearest-neighbor count and the minimum similarity score.
- Added a Processing Errors document field. Documents that failed processing in Case Explorer can be searched and faceted, and the field can be shown as a column in the document list.
- Case Explorer matter settings, such as matter time zone, are now available to Case Explorer.
- Case Explorer chat usage is now reported separately for v1 and agentic (v2) chat. Token counts are broken down by model.
- Fixed the Case Explorer AI document count being over-reported. Documents were counted once per processing stage instead of once per job.
- Export & Relativity Promotion:
- Case Explorer and ECAi fields now export with values when promoting to Relativity. Previously these columns were present in the load file but empty.
- Export templates now support the SingleObject field type
- Export load file generation is faster and uses less memory. It now queries only the data the export template needs.
- The number of child and related documents loaded at a time during export is now capped by a configurable limit.
- Added an optional matter-level setting that speeds up the export fields All Source Folders, All Source Paths, All
- Source Email Folder and All Media ID on large matters. It is off by default.
- The Relativity Server file transfer buffer size can now be configured. Increasing it from the default 16 KB to 1MB showed roughly 43% faster throughput in initial developer testing, which was not a controlled benchmark.
- Relativity promotion now has richer progress logging and metrics (promotion duration, document counts, file and byte counts, per-batch timing and memory usage). This supports monitoring and troubleshooting.
- Fixed Custodian single-object and single-choice fields not being populated on Relativity One export (and checked for Relativity Server).
- Fixed Relativity promotion retry after a failure.
- Ingestion
- Reindexing can now be disabled at the repository level.
- Added an option to route Exodus full-extraction jobs through the Customizable Pipeline for selected repositories. It is off by default.
- Improved logging for Exodus import processing.
- Source deletion processing is more reliable. A single failed deletion no longer blocks the remaining ones, and failures are logged with the task and document source IDs.