Release Notes - 3.04.0

The 3.04.0 release focuses on scaling Rhino for larger, more complex projects while making the Knowledge Reserve, Copilot, Graph Explorer, and artifact workflows more reliable in day-to-day use. It introduces stronger support for multi-asset and path-based routines, improves KR visibility and file handling, hardens Copilot execution on large requests, and delivers major graph and Dynamic Object performance gains for high-volume environments.

 1. Highlights

• Knowledge Reserve: clearer visibility, stronger ingestion, and better app zip handling
• Copilot: more reliable long-running execution, better status feedback, and improved large-project behavior
• Graph Explorer: major scale and query-performance improvements for very large DO datasets
• Introducing Artifacts - Generate custom, interactive, wiki-style descriptions of your application in the language of your business or technical users.
• Routines & Platform: better support for multi-asset and path-based execution, plus security and role hardening

2. Knowledge Reserve

The Knowledge Reserve (KR) continues to serve as Rhino's central project knowledge layer — the place where source material, extracted application content, and configuration assets can be accessed consistently across Copilot, routines, and artifact generation workflows.

• Improved ingestion reliability: fixed cases where non-UTF-8 files were not ingested correctly, expanding the range of source files Rhino can process successfully.
• Better project-start visibility: resolved issues where zip files uploaded during project creation did not appear correctly in KR apps or only appeared in extraction views.
• Exploded app usability: exploded applications are now surfaced more clearly as folders and viewable assets, making uploaded application contents easier to inspect.
• Status feedback: added asset explosion banners and per-asset indicators so users can better understand whether uploaded applications are still being unpacked, completed, or need attention.
• Recovery support: a retry action is available for failed asset explosion scenarios, reducing the need for manual re-upload cycles.
• Safer file separation: fixed cases where matching filenames across steering and supporting documents could overwrite or move the wrong row.
• Cleaner deletion behavior: deleting an app zip no longer leaves exploded child members behind as orphaned assets.
• Copilot refresh behavior: KR updates made through Copilot now refresh more predictably so newly written content is visible without confusion.

3. Documentation & Artifacts

• Documentation generation is more resilient at scale, including support validated against very large Dynamic Object datasets.
• Resolved a prompt-length failure mode that could block model execution on larger requests.
• Fixed User Stories document generation failures seen on specific providers such as Claude.
• Export and generation actions now present more clearly in the UI; corrected cases where document export options appeared disabled.
• Improved reuse of existing DOs during artifact generation, reducing rework in repeated artifact flows.
• Artifact Viewer enhancements add overview pages, improved rail navigation, and richer detail views for Use Cases and functional-spec artifacts.

4. Copilot

• Large-project reliability: improved Copilot behavior for high-scale environments, including fixes for timeouts in graph-backed summary operations on large projects.
• Long-running request handling: added wait/sleep capability so Copilot can pause appropriately while routines complete instead of failing prematurely.
• Clearer runtime feedback: improved visibility when long Copilot requests are still running versus failed, reducing uncertainty for users.
• KR access improvements: fixed failures when reading KR content, including xlsx support issues and Nova-specific “Response Stopped” errors.
• Skill execution hardening: addressed failures where skill execution did not start or where tool call limits were too restrictive for skill-driven flows.
• Chat stability: resolved issues where chat requests and replies could disappear during interaction.
• Routine access parity: restored view and edit behavior for project custom routines in protected-role scenarios.
• Evaluation improvements: expanded the Copilot evaluation framework for skills scenarios and larger customer-like datasets to improve repeatable quality testing.

5. Routines, Extraction, Graph & Observability

• Run on paths: routines now more fully support execution on one or more selected paths, improving precision for targeted analysis.
• Multi-asset execution: expanded support for multi-asset upload and processing, including the ability to run project overview across multiple assets.
• Large-scale reliability: scaled routines for demanding project sizes, including customer-scale scenarios.
• Query optimization: major backend improvements reduce expensive scans in graph nodes, children, search, visualization, and link queries.
• Graph Explorer performance: improved behavior for very large datasets through batching, pagination, and removal of non-scaling patterns such as full-graph visualization.
• Tree view performance: DO instances now load in batches for better responsiveness in large projects.
• Visualization safeguards: graph visualization mode can now be disabled automatically for large projects based on node statistics.
• Platform observability: query-path and scale work lays a stronger foundation for diagnosing performance issues in large environments.

6. Dynamic Objects, Import/Export & UI

• Delete-instance search now scales better beyond very large node counts.
• Improved delete and write speed of DOs within routines, reducing friction in bulk operations.
• Resolved duplicate entries shown in “Delete instances” flows.
• Fixed “No Children” import issues affecting DO imports.
• Resolved failures when exporting all DOs from the graph.
• Improved tree behavior and pagination, including fixes for unpaginated file-category children in Graph Explorer.
• Removed the validate button from the edit DO screen to simplify the editing experience where it was no longer helpful.
• Addressed UI issues such as navbar selected-state behavior while Copilot is open and broken wiki links in summary pages.

7. Platform & Admin

• IP-protected roles: defined and implemented stronger protected-role handling across the platform, including frontend, backend, and supporting specs.
• Provider readiness: completed internal certification work for GPT-5.1 across extraction and Copilot scenarios.
• Open-source model support: added environment support for OSS model configurations, including customer-driven deployment scenarios.
• Security updates: addressed dependency and vulnerability findings across multiple repositories, including fixes related to path traversal, file execution risk, and UUID handling.
• Migration hardening: fixed migration failures tied to user scope and duplicate graph index keys.
• Session handling: improved behavior when sessions expire across multiple tabs.
• Project cleanup: deleting a project now cleans up related graph data more reliably.

Known Behaviors

• Extremely large projects may still use guarded graph behaviors, including disabled full visualization or simplified loading strategies, to preserve responsiveness.
• Some very large skill or artifact flows may still depend on provider-specific limits, but failure handling and user feedback are clearer than in prior releases.