Skip to content
  • There are no suggestions because the search field is empty.

Release Notes - 3.05.0

The 3.05.0 release focuses on making Rhino safer and more execution-ready for real project work through read-only controls, stronger Copilot orchestration, improved Knowledge Reserve editing, spreadsheet generation, and major security hardening. It introduces live Copilot plan visibility, expands MCP and read-only behavior across the platform, improves artifact generation against real project data, and hardens routine, upload, and large-scale execution paths for production use.

Audience: Product owners, solution architects, documentation authors, and admins upgrading from 3.04.0

1. Highlights

  • Read Only projects: broader enforcement across Copilot, KR, Document Studio, and MCP
  • Copilot: live execution plan visibility, stronger tool use, better clarification behavior, and improved model support
  • Knowledge Reserve: editable configuration content, spreadsheet generation, and more consistent asset handling
  • Artifacts: stronger custom artifact generation, editing foundations, and improved real-data acquisition
  • Platform & Security: major dependency remediation, database image hardening, and async-native upload improvements

2. Knowledge Reserve

The Knowledge Reserve (KR) continues to evolve as Rhino’s shared project knowledge layer — the place where source material, editable configuration content, generated knowledge, and reusable skills come together for Copilot, routines, and artifact workflows.

  • Editable configuration workflows: expanded KR editing support in the UI for configuration content, including markdown-oriented editing patterns and dedicated handling for skill-related files.
  • Spreadsheet generation support: introduced production-ready spreadsheet fill and creation capability, enabling Rhino to inspect templates, populate structured workbook outputs, and store generated spreadsheet knowledge back into KR.
  • Generated Knowledge support: KR now more explicitly supports generated spreadsheet-style outputs as a managed knowledge type for downstream use.
  • Bulk and cross-project workflows: improved support for bulk upload and project-to-project knowledge import scenarios in KR release workflows.
  • Consistent exploded asset behavior: fixed inconsistent handling of exploded zip contents so extracted children remain represented more predictably across KR and asset-picker surfaces after parent zip deletion.
  • Supporting document ingestion reliability: addressed gaps where uploaded supporting documents could fail to provide extracted text back to Copilot, improving read and search behavior for formats such as docx, pptx, and pdf.
  • Configuration layout improvements: skills and steering documents now use a more consistent horizontal-card presentation aligned with other KR surfaces.
  • Skill visibility fixes: resolved cases where skills created via Copilot were confirmed as created but did not appear in the visible skill list.

3. Documentation & Artifacts

  • Documentation behavior in read-only projects is more tightly controlled, including disabling create and delete flows where users should not be able to modify content.
  • Custom artifact generation is more robust against real project data instead of relying on incomplete user-provided inputs alone.
  • Artifact creation now uses a fuller page-based experience rather than a more constrained creation flow.
  • Artifact architecture and template language were finalized further, strengthening the skill-based approach for flexible custom artifact definition.
  • Artifact generation now better acquires missing data automatically, reducing the need for users to reason about extraction or schema gaps themselves.
  • Editing foundations for artifact-backed Dynamic Objects were advanced so artifact content can be updated against underlying shared data rather than treated as isolated page text.
  • Security around artifact creation and editing was strengthened through safer execution, validation, and access-control hardening.
  • Older artifact compatibility was restored for legacy templates affected by stricter content security policy changes, preventing breakage of existing rendered artifacts.

4. Copilot

  • Live plan visibility: Copilot now shows a user-visible plan or todo checklist during execution so users can see the steps being worked, their status, and progress over time.
  • Clarification behavior: improved prompting behavior so Copilot asks focused follow-up questions when user input is incomplete instead of overcommitting to ambiguous requests.
  • Automatic gap resolution: Copilot more consistently attempts to obtain missing project data itself before producing outputs such as artifacts.
  • Read-only behavior: write attempts in read-only projects are handled more appropriately, including suppressing misleading workaround guidance and respecting project restrictions across more flows.
  • Prompt and tool-use quality: ported stronger anti-redundancy, anti-fabrication, counting, scoping, and tool-responsibility guidance into the newer Copilot execution path.
  • MCP and tool parity: improved MCP behavior for read-only contexts and strengthened compatibility between Copilot and MCP-backed project operations.
  • Model support: added support for Claude Fable 5 in Copilot and improved handling for reasoning-model edge cases across OpenAI and Bedrock-backed configurations.
  • Error handling: fixed multiple Copilot failure modes, including recurring internal errors on some model deployments, invalid tool-result sequencing, and cases where codebase tools retried instead of self-correcting.
  • Stability on busy environments: mitigated intermittent 502 failures caused by long-running Copilot requests and infrastructure health-check interactions.
  • Evaluation framework: improved the Copilot evaluation harness so it works more reliably across different projects and models, giving better benchmarking and regression coverage.

5. Routines, Extraction, Graph & Observability

  • Graph-based routine execution: routine runs no longer require file paths or asset IDs when the routine operates on graph-backed data instead of source-file targets.
  • Large reduce stability: improved long-running reduce-block behavior on Bedrock-backed routines by moving critical synthesis flows to streaming-friendly execution paths.
  • Prompt-size resilience: addressed large-prompt failure cases seen in demanding routine scenarios such as complex use-case and journey synthesis.
  • Tool-result sequencing fixes: resolved validation failures in routines caused by missing or mismatched tool-result blocks on some model/provider combinations.
  • Routine history performance: the routine history view now avoids loading full Dynamic Object payloads when only summary information is needed, improving load behavior on large projects.
  • Status visibility: restored clearer status reporting while waiting on routine execution, including better visibility into running, failed, and cancelled work.
  • Routine detail visibility: project-scoped custom routine definitions are now viewable for non-power-users where appropriate, while system routine protections remain intact.
  • Upload and async modernization: asset upload handling was moved toward async-native execution with FastAPI and async database access, reducing WSGI bottlenecks and improving scalability foundations.
  • Observability and execution architecture: continued work on the planning and execution layer behind complex skill-driven automation, creating a stronger base for repeatable multi-step routine and Copilot workflows.

6. Dynamic Objects, Import/Export & UI

  • Resolved “LLM not configured” and related model-construction failures when creating new Dynamic Object definitions from the UI on certain model setups.
  • Fixed DO generation failures on OpenAI reasoning models where function tools and reasoning settings conflicted with the wrong API path.
  • Improved node deletion flows driven through Copilot so confirmation behavior is more reliable.
  • Advanced support for editing artifact-backed underlying DO content rather than only treating generated pages as static output.
  • UI consistency improved across read-only indicators, KR cards, project-list tagging, and other cross-surface states.
  • Routine- and graph-related UI flows are more predictable for larger projects and higher-volume Dynamic Object datasets.

7. Platform & Admin

  • Read-only project enforcement: implemented and refined broader read-only controls across KR, Document Studio, Copilot, project settings, and related admin/user views.
  • Security hardening: completed major remediation work across frontend, backend, transformer, and dependency surfaces, addressing numerous high- and medium-severity vulnerabilities in packages such as DOMPurify, undici, ws, form-data, vite, cryptography, markdown-it, nltk, pypdf, and others.
  • Database image hardening: advanced the 3.05 security program to rebuild the PostgreSQL + Apache AGE image from fresher bases, reducing inherited CVE load and improving control over patch cadence.
  • Migration reliability: improved behavior for expensive database migration steps such as trigram indexing so upgrades are less likely to timeout at production scale.
  • Developer and deployment updates: updated Azure configuration defaults for local deployment tooling and continued refactors that move shared auth and DB adapter infrastructure into more appropriate common services.
  • Custom deployment support: improved support for deployment-specific KR assets and plugin-style startup content in custom environments.

Known Behaviors

  • Some of the newest spreadsheet and artifact workflows are more capable than prior releases, but quality can still vary based on the structure and complexity of the source templates or project data.
  • Very large graph, routine, and artifact scenarios may still rely on guarded execution strategies to preserve responsiveness and avoid provider or infrastructure limits.
  • Legacy artifact compatibility has been preserved for this release, but older rendering paths remain transitional and may be retired once replacement flows are fully adopted.