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

Release Notes - 3.07.0

Rhino 3.07.0 builds on the Knowledge Reserve, artifact, Copilot, task execution, and security hardening delivered in 3.06.0. This release focuses on making generated knowledge more useful in real work, improving Knowledge Reserve retrieval and visibility, expanding routine and Copilot execution patterns, and closing a broad set of reliability issues found during release testing across Copilot, MCP, artifacts, tasks, and platform operations.

Audience: Product owners, solution architects, documentation authors, admins, and technical teams upgrading from 3.06.0.

1. Highlights

  • Knowledge Reserve and Generated Knowledge improvements: text deliverables, folder support, in-app preview, stronger search behavior, enhanced indexing for Supporting Documents, and better handling of project-scoped and system-scoped content.
  • Routines and execution flows: script routines can now write files directly to the Knowledge Reserve, with stronger custom-routine testing and execution reliability.
  • Artifacts and scale: faster artifact rendering, large-scale Dynamic Object handling, export support, direct filter controls, and stronger sandbox error feedback to Copilot.
  • Copilot and task hardening: better recovery from failures, fewer leakage and streaming issues, improved task execution workflows, and stronger built-in skill availability and project-scoped behavior.
  • Security and operations follow-through: continued CVE remediation, SBOM and provenance work, cost visibility, and automation improvements around vulnerability handling and release support.

2. Knowledge Reserve

The Knowledge Reserve continues to mature as Rhino’s central layer for project knowledge, generated outputs, supporting documents, and reusable system content.

  • Generated Knowledge now supports text deliverables: Copilot and system workflows can write Markdown, text, CSV, JSON, YAML, and HTML outputs in addition to existing spreadsheet generation.
  • Folder-based organization for generated outputs: generated files can now be written into nested folder paths instead of a single flat list, making multi-file outputs easier to manage.
  • In-app preview for supported text formats: users can open generated Markdown, text, CSV, JSON, YAML, and HTML files directly in the product, while binary outputs remain download-first.
  • Search and navigation improvements: release testing validated search bar behavior, progressive loading, folder-based navigation for content buckets, and better visibility into large Knowledge Reserve structures.
  • Global and system-scoped content usability: Copilot can better work with globally scoped Knowledge Reserve content, and UI behavior better reflects non-editable system-scoped files.
  • Enhanced Smart Search for Supporting Documents: Postgres Full Text Search, ranking on body text in addition to metadata.
  • Import and project-path fixes: imported Knowledge Reserve files now align more reliably with the target project’s scoped paths so they remain visible in UI and Copilot flows.
  • Best-fit retrieval exploration: additional work evaluated query expansion and document-selection strategies for best-fit supporting-document retrieval, building on the content-focused search direction established previously.

These changes make the Knowledge Reserve more practical as both a working knowledge layer and a generated-output destination, especially for larger projects and document-heavy workflows.

3. Artifacts, Custom Applications, and Embedded Experiences

Artifacts continue evolving from generated outputs into interactive project surfaces that must scale, render quickly, and provide strong authoring feedback.

  • Artifact performance hardening: improvements reduced time-to-first-render and strengthened data access for very large Dynamic Object sets, including support targets at 500k+ objects per project.
  • Direct filter controls: artifact detail pages now support direct UI-based filter changes, reducing dependency on Copilot for simple slice-and-view interactions.
  • Artifact export support: artifacts can now be exported more directly, improving downstream sharing and delivery workflows.
  • Render-error feedback to Copilot: browser-side sandbox render failures, including Mermaid-related issues, can now be surfaced back into Copilot so the agent can self-correct on later turns.
  • Sandbox and typed-item fixes: follow-up fixes addressed child-data loss in sandbox artifacts after scaling changes and improved launch behavior so sandbox-initiated builds open correctly in Copilot.
  • Viewer navigation and usability fixes: artifact close behavior and related navigation paths now behave more consistently, especially when artifacts are opened directly by URL.

Rhino is increasingly treating artifacts as a living application and review surface, not just a static deliverable layer.

4. Tasks, Routines, and Project Execution Flows

3.07.0 advances the connection between tasks, routines, Copilot, and file-generating workflows.

  • Routine write-to-KR capability: script routines can now write text files and spreadsheet outputs directly into Generated Knowledge, reducing the need to rely on Copilot for large-scale file generation tasks.
  • Generated Knowledge as an execution target: Copilot-driven workflows now default more naturally toward Generated Knowledge for user-facing deliverables such as reports and exports.
  • Task model improvements: work completed in this release strengthened task typing, execution semantics, dependency handling, and the groundwork for more reliable agentic-task execution.
  • Re-run and recovery improvements: users can more easily re-run edited tasks, and several fixes addressed stuck, blocked, or mismatched task/run states.
  • Optional and generic skill support: agentic tasks can more easily work in environments where a bespoke skill has not yet been authored.
  • Routine and custom authoring separation: Python routine authoring controls were refined so human authoring and Copilot authoring can be managed separately and more safely.
  • Better routine execution validation: release work included stronger testing and fixes for custom script routines, import/export flows, and concurrent routine-run reliability.

5. Copilot, MCP, and Reliability Hardening

A major focus of 3.07.0 is continued hardening of Copilot behavior across chat, skills, retrieval, task execution, and model-provider differences.

  • Failure isolation and recovery: fixes addressed scenarios where one failed turn could poison later turns, improving Copilot’s ability to recover cleanly instead of cascading repeated failures.
  • Reduced internal-state leakage: work closed cases where internal summaries, scratchpad text, raw todo data, or reasoning-like content could leak into user-visible answers.
  • Improved tool-path robustness: fixes addressed glob crashes on archive and absolute paths, opaque Knowledge Reserve smart-search failures, regex grep crashes, pagination mismatches, and other tool-level regressions.
  • Built-in skills availability fixes: built-in Copilot skills now surface correctly to non-power-users on the affected architecture path, improving discoverability and execution reliability.
  • Streaming and persistence improvements: fixes addressed dropped preamble text, duplicated reasoning blocks, misplaced persisted content ordering, stale draft-chat handling, and chat-deletion edge cases.
  • Approval and HITL flow hardening: approval-gated skills, autopilot behavior, and input-request rendering were improved so users receive clearer context and fewer dead-end execution states.
  • Task and chat behavior consistency: project-scoped chats, project access, run launches, and view-chat flows behave more predictably across teams and shared execution contexts.
  • MCP and Bedrock follow-up fixes: release work included MCP tool-path validation and Bedrock-specific fixes, including improved compatibility for Boomi and better reasoning-effort handling on non-Copilot model paths.