Ingestion and conversation uploads

Source ingestion prepares documents for retrieval. Conversation uploads attach files to an individual chat. They have different configuration and authorization requirements.

Choosing where the orchestrator runs does not choose either path. An agent strategy controls orchestration behavior, not whether an ingestion service is deployed.

Choose a source ingestion path

Path	What processes the source	When to use it
Native Foundry IQ Blob source	Azure AI Search processes the configured `azureBlob` knowledge source	Your documents and permission model fit the supported native Blob configuration. No ingestion Container App is required for this path.
Independently populated Search index	Your own ingestion pipeline populates a compatible index	You bind `SEARCH_RAG_INDEX_NAME` explicitly, using `ai_search` or Foundry IQ `searchIndex`. Disabling the bundled app does not create this pipeline or populate the index.
Agent App Ingestion	The optional ingestion Container App processes documents and writes Search content	You need the application's processing pipeline, its SharePoint jobs, or the conversation-upload path. Review the selected component's supported formats and configuration.

Agent App Ingestion provides format-specific chunking, text/image embeddings and application-managed ingestion jobs. It is not simply a deployment wrapper for native Foundry IQ ingestion. Neither path is a drop-in replacement for the other.

SharePoint is a separate integration

The optional application's Graph-based SharePoint jobs and Foundry IQ's preview indexedSharePoint knowledge source are different implementations. Configuring native Blob ingestion does not configure SharePoint.

The platform includes a preview indexed SharePoint registration template, but its payload is not compatible with the current documented API contract. Supplying identifiers and consent is not enough to make it a working integration. See SharePoint grounding for the supported service setup, the hosted integration requirements and the distinction from application-managed Graph jobs.

The bundled ingestion opt-out validator supports native azureBlob, or an explicit independently populated index. It does not establish SharePoint-only ingestion-free support or parity with the application's SharePoint jobs. Validate the chosen integration and its document authorization before removing an existing ingestion dependency.

Select the optional application

DEPLOY_INGESTION_CONTAINER_APP accepts only true or false:

Environment	Resolution
New bundled application	Defaults to `false`.
Existing bundled application	Retains its recorded selection. A populated older environment without a selection retains ingestion. A failed state lookup requires restored access or an explicit choice, not a guess.
Explicit custom definition	Retains its declared component intent and the current additive foundation composition. The bundled-only switch does not remove its components.

An empty azd-created resource group is considered fresh. Pre-provision persists the resolved choice, and configuration publishes the effective value under App Configuration label agent-lz.

Contextual setting, applied before the matching deployment runbook:

azd env set DEPLOY_INGESTION_CONTAINER_APP false

For the bundled application, opt-out excludes ingestion from subsequent provisioning, checkout, build, deployment, endpoint reconciliation and identity/role reconciliation. It does not change the retrieval backend, source/index bindings, strategy, topology or document authorization.

An ingestion-free ai_search deployment needs SEARCH_RAG_INDEX_NAME. For Foundry IQ with an independent index, also select both FOUNDRY_IQ_PATTERN=searchIndex and FOUNDRY_IQ_KNOWLEDGE_SOURCE_KIND=searchIndex. The index must pass the existing schema/upgrade guards. Unsupported combinations, including FOUNDRY_IQ_CONVERSATION_UPLOAD_ENABLED=true with ingestion disabled, fail validation.

Conversation uploads are not native source ingestion

The bundled UI reads the effective ingestion flag:

false hides attachment controls and rejects server uploads and client preflight before ingestion endpoint or Dapr fallback resolution.
true retains upload eligibility checks. It does not give every caller permission to upload.
Missing retains legacy behavior, not a safe opt-out.

Text chat, delegated authentication and retrieval remain available when uploads are disabled. When enabled, authenticated-mode eligibility, anonymous Copilot restrictions, conversation ownership/eligibility, supported file types and size/count limits still apply. Hosted upload eligibility also requires a conversation identifier. Native source ingestion does not implement this chat attachment path.

Custom applications own their upload contract. A published capability flag does not prove that a custom application implements uploads.

Upgrade and retirement safety

Before changing an existing selection, record the current platform revision, manifest, source/index bindings and ingestion dependencies. Back up relevant configuration and data. Apply selection changes through the full deployment workflow, not deploy-only or a manually edited runtime key. Private environments need a connected runner.

Older UI compatibility

The upload gate was introduced in the coordinated Agent Landing Zone v4.3.0 / agent-app-ui v3.3.0 combination. On v4.2.3 / UI v3.2.2, retain explicit DEPLOY_INGESTION_CONTAINER_APP=true: that UI sends attachments to ingestion in both hosting modes and does not enforce the new gate. An empty endpoint can fall back to a derived URL or Dapr. Changing Chainlit TOML or authentication policy is not a supported opt-out.

Upgrade the complete manifest-pinned combination before disabling ingestion. Check that App Configuration publishes the resolved flag and the deployed UI consumes it. Verify attachment controls are hidden and stale/direct uploads are rejected, while sign-in, chat, retrieval and document allow/deny tests still work.

Opt-out does not delete an existing app, documents or indexes, or revoke existing identity permissions. Resource retirement and RBAC cleanup require separate, deliberately authorized actions after validation.

To restore ingestion, restore the previous explicit selection and source/index configuration, rerun the matching full workflow and retest. Enabling the app alone does not repair bindings. Before rolling back to a UI without the upload gate, restore ingestion and its required endpoint/source configuration first. Rollback must not delete documents or indexes.

Authorization remains a design requirement

Preserve authentication, Entra ID/OBO, RBAC and document-level authorization across ingestion and retrieval. Test with both an allowed user and a user who must not see the document. A successful indexer, greeting or hook is insufficient.

The default native Blob permission metadata is ["rbacScope"]. An explicit [] is a deliberate shared-source choice only for supported active non-ADLS native Blob ingestion, not a private-link or login repair. See permission options and private Blob connectivity.