v3.3.1

Fixed

• Deploy Container Apps, registry operations and deployment identities to AZURE_APP_RESOURCE_GROUP from Azure App Configuration under agent-lz.
• Preserve the legacy AZURE_RESOURCE_GROUP fallback only for confirmed missing or empty application-group settings. Configuration access failures stop deployment.
• Keep host deployment-script tests separate from installed runtime acceptance.

Validation

• Source unittest suite, non-editable installed-wheel acceptance and ephemeral Linux-container acceptance.
• Protected lint, typing, architecture, exception, policy and aggregate quality checks.

v3.3.0

Changed

• Optional-ingestion upload capability. Honor DEPLOY_INGESTION_CONTAINER_APP independently of the chat backend: false hides standalone and Entra Copilot upload controls and rejects attachments before ingestion URL or Dapr resolution.
• Missing preserves legacy upload eligibility; true retains existing authentication and conversation checks; malformed boolean values fail with a configuration error. Anonymous Copilot uploads remain disabled.
• Text chat and retrieval are unchanged; no native Foundry IQ conversation upload path is introduced.
• Component README and embedding documentation describe the capability and its compatibility behavior.

Validation

• Offline only; no Azure acceptance claim. No live Azure checks or reads, Azure deployments, cloud builds, environment changes, or manually triggered deployment workflows were performed.
• Feature PR Azure/agent-app-ui#151 and release PR Azure/agent-app-ui#152 passed the existing CI without bypasses or quality-policy changes.
• python .github/scripts/run-unittest.py --base-ref <protected-base-sha> --report .artifacts/unittest.json: 576 source tests passed, including 11 non-editable installed-package acceptance tests and their 506 installed-wheel behavioral tests.
• Ephemeral Linux-container acceptance: 506 behavioral tests passed offline, plus installed-origin, staged-resource, and startup checks. No image or package was published.
• Protected lint, typing, architecture, exceptions, and policy checks each passed with zero findings; aggregate quality-gate passed.
• python .github/scripts/validate-agentic-assets.py: passed.
• Release CI and source-bound JSON evidence.
• Exact released-main CI and source-bound JSON evidence passed at 858ed4205d866b7bfbff43ca9ed20ab812baa24e.
• The release changes only VERSION and CHANGELOG.md relative to the validated feature tree. The released main tree is identical to the validated release branch.

Compatibility and documentation

• Backward-compatible UI minor release using an existing configuration capability; no new backend wire contract. A missing setting preserves legacy behavior.
• This is a UI source release, not certification of a new platform-wide pinned combination. Central platform documentation and manifest adoption remain in the subsequent coordinated platform/docs change.
• VERSION is 3.3.0; CHANGELOG.md records v3.3.0 dated 2026-10-14 and retains exactly one empty Unreleased section according to UI conventions.

Rollback

• Consumers may retain or restore the previous UI pin v3.2.2 through the reviewed platform process. No deployment is performed by this release.
• Preserve published tags and history; use a reviewed corrective release for source changes rather than moving or replacing this tag.

v3.2.2

Changed

• Include api-version=v1 in Hosted invocations and support both the current response.conversation.id and legacy response.conversation_id envelopes, rejecting malformed or conflicting IDs.
• Register classic Dapr-backed conversation history only for the orchestrator backend. Hosted context remains session-scoped; OAuth, feedback and classic authenticated history are preserved.
• Incorporate the previously reviewed technical exception-registry validation policy without weakening source-bound evidence, expiry, protected tools or fail-closed aggregation.

Validation

• Installed-wheel unit acceptance and the offline Linux container checks passed.
• A genuine post-adoption reference pull request passed lint, typing, architecture, exceptions, policy and the aggregate quality gate against the adopted protected main evaluator.
• Azure environment validation and deployment were intentionally omitted at the maintainer's request.

v3.2.1

Changed

• Upgrade Chainlit to 2.12.0, aiohttp to 3.14.3, and setuptools to 83.0.0 to resolve Dependabot security alerts; add the get_favorite_steps data-layer callback required by Chainlit 2.12.

Validation

• 550 UI unit tests and 54 quality-policy tests passed; all five quality checks passed with zero findings.

v3.2.0

Removed

• Legacy gpt-rag App Configuration label fallback and GPT_RAG_ key fallback (Azure/agent-landing-zone#695). The app now reads only its app label, then agent-lz, then no label.

Validation

• Unit test suite passed locally and in CI.

v3.1.1

Fixed

• Fail early with a clear error when NETWORK_ISOLATION=true, BUILD_MODE=acr-task and ACR_TASK_AGENT_POOL_NAME is missing (Azure/agent-landing-zone#741).

v3.1.0

Added

• Keyless service-to-service auth to the orchestrator (ADR-0019). The UI
  acquires a Microsoft Entra token with its managed identity (scope from
  ORCHESTRATOR_AUTH_AUDIENCE) and sends it in X-Service-Authorization,
  keeping Authorization for the user's OBO token. API key and Dapr remain
  opt-in fallbacks.

v3.0.0

Changed

• Agent Landing Zone dual-read and rebrand (Azure/agent-landing-zone#695). App
  Configuration now loads the agent-lz label, which overrides the legacy
  gpt-rag label; AGENTLZ_* keys are read before legacy GPT_RAG_* keys.
  Tracer names use the agentlz. prefix. Visible branding (assistant name,
  welcome title, release footer) now reads "Agent Landing Zone". Existing
  gpt-rag deployments keep working unchanged.

Fixed

• Retry failed classic-chat uploads without losing attachments. Restore the
  previous session conversation ID when a failed upload allocated a new one.
  Reattached files are ingested under a fresh conversation before the question
  runs; existing conversation ownership checks remain unchanged.
• Stop questions after unconfirmed ingestion. False results and raised
  failures finalize a reference-bearing retry notice without sending the
  question or claiming successful file processing. Only confirmed batches
  update uploaded-document bookkeeping.
• Keep OpenAPI failures retryable and truthful. Failed schema generation
  propagates instead of returning a successful metadata-only schema; later
  requests retry generation and cache only success.
• Expose owner-index repair needs without weakening panel access. A failed
  optional index write logs that repair is required while the turn continues;
  missing ownership rows still hide panel listings and deny read, feedback,
  and delete operations.

Changed

• Preserve dependency failures without masking implementation defects.
  Narrow configuration, HTTP/JSON, Blob, JWT, logging and VERSION handlers to
  their concrete failure contracts. Fix the unbound retry notification and
  avoid retrying absent in-memory settings. Record retained application
  boundaries individually as unapproved proposals with failure-test evidence.

• Separate Chainlit history adaptation from history operations. Put the
  data-layer callbacks, factory and consume-once session bridge in api.history;
  keep history authorization, orchestrator operations and the single user cache
  in services.history with explicit context. Preserve legacy exports and
  ownership-before-session-selection ordering without a storage rewrite.

• Canonical UI runtime package with legacy adapters. Move runtime ownership
  into src/gpt_rag_ui, share citation rendering between chat and restored
  history, and separate callback registration from reusable operations. Keep
  uvicorn main:app, existing configuration/defaults and externally staged
  Chainlit assets; install the package in the existing container build.

Added

• Isolate protected quality-tool execution. Run static tools without
  candidate module shadowing or candidate package/build-backend installation
  in the evaluator; analyze the explicit source graph in an isolated worker.
  Add executable shadowing and installation-isolation regression fixtures.

• Proposed protected-base Python quality gates. Add pinned development-only
  Ruff, mypy and import tooling, individual typing debt records, full static
  dependency checks, an explicit broad-handler inventory and a fail-closed
  quality-gate aggregate alongside unittest. Initial handler approvals,
  bootstrap review and administrator activation remain outstanding; this
  change does not claim required merge enforcement is active.

• Exact quality policy and executed failure evidence. Validate closed
  governance schemas, stable move/debt identities, annotation/suppression sites,
  adapter exports and dynamic imports. Bind exception approvals to distinct
  protected operations and same-run, non-skipped unittest results. Require
  protected aggregation and ephemeral offline Linux-image acceptance without
  publishing an image or changing repository settings.

• Close binding and namespace quality-check bypasses. Detect qualified
  suppression decorators on untyped functions, invalidate exception records
  when effective catch bindings change, and reject unresolved catches.
  Discover root namespace modules and retain their full typing/import identity
  without changing application imports or exception behavior.

v3.0.0-preview.2

Changed

• Deploy scripts read App Configuration settings from the agent-lz label with fallback to the legacy gpt-rag label.

Pre-release for Agent Landing Zone v4.0.0-preview.2.

v3.0.0-preview.1

Changed

• Agent Landing Zone rebrand: dual-read of AGENTLZ_* settings with temporary gpt-rag fallback.
• Audit telemetry prefix agentlz.audit.* (contract v2).

Pre-release for Agent Landing Zone v4.0.0-preview.1.