v0.8.0

[0.8.0] - 2026-09-10

Added

• Initial loading skeletons now match the layouts of Assets, Dashboard, Builds, Runs, Improvements, and Settings.
• Runner workflow graphs now annotate lifecycle steps with execution information.
• The runner lifecycle and sortable-list controls are generalized for reuse across workflows and catalog views.
• Direct dependency license inventory is documented in the repository.

Changed

• Evaluation Builds are now called Builds, and Improvement Results are now called Improvements, throughout the application, documentation, and localized UI.
• Workflow lifecycle phases are standardized as before_all, before_each, execute, verify, after_each, and after_all.
• Application-data path editing and localized help keys have clearer, more stable behavior.

Breaking changes

• Build API routes now use /api/builds instead of /api/evaluation-builds; clients must update build create, read, update, delete, run, and test requests.
• New workflow definitions and run records use the standardized lifecycle phases and build_id / build_name fields. Legacy lifecycle phase names remain readable for stored workflows and integrations.

v0.7.0

[0.7.0] - 2026-09-10

Added

• The Orbit Assistant can use workspace tools and an optional terminal mode, with configurable tool availability for local operation.
• Evaluation runs now retain structured supervisor behavior traces, target logs, live phase output, retry controls, scheduled execution settings, prompt revision history, and commit-change evidence.
• Evaluation result, workflow, log, and supervisor views now expose contextual data hints, line-numbered output, richer filtering, and independent display-only translation state.
• Runner SDK documentation is bundled with the release, and SDK helpers can restore evaluation targets from retained Git snapshots and surface saved data files in proposal history.
• Assets now show loading skeletons, reusable execution and target environments, and clearer catalog metadata and actions.
• Improvement results include separate dashboard feedback trends and per-build iteration improvement trends, proposal history data files, and localized operational analytics.
• The control room now provides detailed section descriptions and contextual tooltips across dashboard, evaluation, assets, and settings pages.

Changed

• Evaluation-build scheduling, timing controls, and date/time displays consistently follow the selected Orbit locale.
• Operational logs now format timestamps with the selected Orbit language rather than the browser default.
• The operational manager prompt, application-data settings, model-profile settings, and asset catalogs have clearer descriptions, warnings, and aligned layout spacing.
• The bundled frontend release label is verified against the package version during CI, and generated SDK documentation is included in release build artifacts.
• UI accessibility and static copy are localized consistently across English, Korean, and Japanese.

Fixed

• Evaluation detail hints remain visible in the run-detail modal, and asset metadata aligns with catalog-row actions.
• Model profiles can be saved without a connection test when the profile configuration is otherwise valid.
• Evaluation-build timing controls and retained evaluation detail data are localized correctly.

v0.6.0

[0.6.0] - 2026-09-08

Added

• Agent self-improvement Quick Starts can now evaluate real target-AI responses through the selected model profile and retain response evidence as a run artifact.
• The runner SDK now provides a credential-safe target-AI completion helper for evaluation assets.
• The README includes a ten-second Quick Start guide, representative templates, and a response-level prompt-improvement example.

Changed

• The managed prompt sample is intentionally minimal so the response-evaluation Quick Start can demonstrate evidence-backed improvement from a real behavior gap.

Fixed

• Evaluation-run history totals now retain proposed and adopted improvements and reported issues from every supervisor iteration, rather than showing only the latest iteration.
• Evaluation-run translations now apply only to the active Supervisor AI iteration or currently filtered result records, with independent loading state for each view.

v0.5.1

Fixed

• Frontend release labels now use the packaged release version, so a release build cannot retain an earlier Git tag.
• Supervisor review now includes setup-phase managed-prompt evidence, allowing directly evidenced prompt-only improvements to be adopted and applied on the next iteration.

OpenOrbit v0.5.0

Added

• Supervisor results include concise evaluated-AI behavior summaries and translated display.
• Manager prompt templates retain version history, preserve drafts, and create a new version per save.
• The dashboard surfaces Quick Starts and links to the repository, release notes, and local OpenAPI docs.

Changed

• Supervisors can adopt evidence-backed, low-risk, reversible prompt-only improvements; higher-risk or insufficiently evidenced changes remain proposed.
• Improvement analytics and dashboard charts now use localized labels, themed tooltips, and preserve short active runs.
• Proposal history, workflow logs, evaluation results, and Cycle Improvement AI have clearer, more compact presentation.

Fixed

• Restored AI model profiles in Assets and aligned Settings with the shared profile component.
• Workflow log output preserves separate log lines.

v0.4.1

Changed

• Evaluation Builds and evaluation execution history now share the same pagination component, layout, and range display.

v0.4.0

Added

• Runner-template and Quick Start catalogs can translate all visible template metadata at once, with cached display-only translations and an option to restore the originals.
• Added icons to catalog translation/import actions and the Evaluation Build duplicate action for clearer controls.

OpenOrbit v0.3.0

Added

• Evaluation run detail now records and displays timestamps for individual log lines; the Logs tab presents one continuous output stream while Workflow logs retain phase-specific inspection.
• The README now includes a product tour with English demo data, standalone/BYOA guidance, Chat Assistant usage, development-partner attribution, and Git-based installation instructions.
• Source and Git installations build the bundled control-room UI into the wheel; Python CI reuses the verified frontend artifact.

Changed

• OpenOrbit is positioned as a local control plane for continuously evaluating, supervising, and improving AI systems.
• Evaluation Builds now display localized created and last-started timestamps.
• Interface text is consolidated through locale resources, including evaluation-run filtering and status labels.

Fixed

• Failed runner phases now stop evaluation runs and correctly mark them as failed instead of completed.
• Evaluation Build selection and the Last started column are restored.

v0.2.0

[0.2.0] - 2026-09-07

Added

• Declarative Quick Start packages can now create reusable evaluation assets and a configured evaluation build from a guided setup flow.
• The Evaluation Builds test action opens a live run-detail dialog with execution status and logs, while test sessions remain out of Evaluation runs history.
• The sidebar displays the current number of queued, approval-pending, and running evaluation runs, capped at 99+.

Changed

• Local evaluation builds and workspace browsing may use any existing local directory; the prior fixed workspace-root restriction was removed.

Fixed

• Evaluation runners now expand home-directory paths such as ~/projects/example before checking that the working directory exists.
• Release lockfile verification now includes the project version, preventing uv sync --locked failures after a version bump.

OpenOrbit v0.1.0

Highlights

• Added filtered, all-iteration evaluation results with clearer evidence and logs.
• Added build-level Cycle Improvement AI with a proposal-decision tree and System AI Markdown diagnosis.
• Derived proposal history directly from evaluation-run supervisor results.
• Improved localization, tooltips, navigation, runner templates, and evaluation forms.

See CHANGELOG.md for the complete release notes.