Azure.AI.Projects.Agents Namespace

An agent implementing the A2A protocol.

An A2A tool stored in a toolbox.

Configuration specific to the A2A protocol.

An A2A tool stored in a toolbox.

Configuration specific to the activity protocol.

The AgentAdministrationClient sub-client.

Client options for Azure.AI.Projects.Agents.InternalProjectsClient.

Represents the settings used to configure a AgentAdministrationClient that can be loaded from an IConfigurationSection.

Represents the settings used to configure a AgentAdministrationClient that can be loaded from an IConfigurationSection.

The AgentBlueprintReference. Please note this is the abstract base class. The derived classes available for instantiation are: ManagedAgentIdentityBlueprintReference.

The AgentCard.

The AgentCardSkill.

The AgentEndpointAuthorizationScheme. Please note this is the abstract base class. The derived classes available for instantiation are: BotServiceAuthorizationScheme, BotServiceRbacAuthorizationScheme, BotServiceTenantAuthorizationScheme, and EntraAuthorizationScheme.

The AgentEndpointConfiguration.

Multipart request body for creating a new code-based agent (POST /agents). Inherits from CreateAgentVersionFromCodeContent for future extensibility.

A managed runtime and agent loop used to execute a prompt agent. Please note this is the abstract base class. The derived classes available for instantiation are: GitHubCopilotHarness.

The AgentIdentity.

The AgentManifestOptions.

The AgentObjectVersions.

Aggregated evaluation result for a single candidate agent configuration across all tasks.

Evaluation criterion: a name + instruction pair used for per-item scoring.

Base discriminated model for dataset input. Either inline items or a registered reference. Please note this is the abstract base class. The derived classes available for instantiation are: AgentOptimizationInlineDatasetInput and AgentOptimizationReferenceDatasetInput.

A single item in an inline dataset.

Reference to a named evaluator, optionally pinned to a version.

Inline dataset — items supplied directly in the request body.

Agent optimization job resource — a long-running job that optimizes an agent's configuration (instructions, model, skills, tools) to maximize evaluation scores. On success, the result contains scored candidates.

Caller-supplied inputs for an optimization job.

Slim job representation returned by the LIST endpoint.

In-flight progress; only populated while status is queued or in_progress.

Terminal-state result body. Populated when status is succeeded or failed.

The AgentOptimizationJobs sub-client.

Tuning knobs and run-mode for an optimization job.

Reference to a registered Foundry dataset.

Session-scoped file operations for hosted agent sandboxes. These endpoints enable uploading, downloading, listing, and deleting files within a live ADC sandbox session. All file content is streamed end-to-end (no buffering in Agents service memory). Sessions are scoped to agent endpoints (not individual versions), allowing the endpoint to evolve the backing agent version independently of the session lifecycle.

A skill resource.

The AgentToolboxes sub-client.

JSON metadata for code-based agent operations (create, update, create version). The agent name comes from the URL path parameter or the x-ms-agent-name header, so it is not included in this model. The content hash (SHA-256 of the zip) is carried in the x-ms-code-zip-sha256 header.

Context class which will be filled in by the System.ClientModel.SourceGeneration. For more information https://github.com/Azure/azure-sdk-for-net/blob/main/sdk/core/System.ClientModel/src/docs/ModelReaderWriterContext.md

The input definition information for an Azure AI search tool as used to configure an agent.

An Azure AI Search tool stored in a toolbox.

A AI Search Index resource.

A set of index resources used by the azure_ai_search tool.

The structure for keeping storage queue name and URI.

The definition of Azure function.

The AzureFunctionDefinitionFunction.

The structure for keeping storage queue name and URI.

The input definition information for an Azure Function Tool, as used to configure an Agent.

Read-only, endpoint-scoped retrieval of the conversations, responses, transcript items, per-turn metrics, and audio recorded by a voice agent. For a voice agent, the service is the sole writer of responses and items, so only reads plus a cascading conversation delete are exposed — there are no response/item create, update, or delete routes. All persistence is governed by the definition's store flag: when store = false (the default) nothing is persisted, so these routes return 404. The merged-recording routes additionally require the session to have ended and return 409 while it is still in progress.

The BetaVoiceAgentsTelephony sub-client.

A bing custom search configuration.

The input definition information for a Bing custom search tool as used to configure an agent.

The bing custom search tool parameters.

Search configuration for Bing Grounding.

The bing grounding search tool parameters.

The input definition information for a bing grounding search tool as used to configure an agent.

The BotServiceAuthorizationScheme.

The BotServiceRbacAuthorizationScheme.

The BotServiceTenantAuthorizationScheme.

The input definition information for a Browser Automation Tool, as used to configure an Agent.

A browser automation tool stored in a toolbox.

Definition of input parameters for the connection used by the Browser Automation Tool.

Definition of input parameters for the Browser Automation Tool.

A tool for capturing structured outputs.

Extension methods on ClientConnectionProvider that produce Foundry-project-specific agent clients.

Code-based deployment configuration for a hosted agent.

A code interpreter tool stored in a toolbox.

Container-based deployment configuration for a hosted agent.

Configuration for Responsible AI (RAI) content filtering and safety features.

The CreateAgentVersionFromManifestRequest.

The request to create a Microsoft Teams Phone Extension binding.

The request to create a telephony binding. Please note this is the abstract base class. The derived classes available for instantiation are: CreateTeamsPhoneExtensionTelephonyBindingContent and CreateTwilioTelephonyBindingContent.

A request to create one durable direct outbound call job.

The request to create a Twilio binding.

The prompt agent definition.

The EntraAuthorizationScheme.

The external agent definition. Represents a third-party agent hosted outside Foundry (for example, on GCP or AWS). Registration is metadata-only: Foundry records the agent definition to light up observability experiences (traces, evaluations) over customer-emitted OpenTelemetry data.

The fabric data agent tool parameters.

A FabricIQ tool stored in a toolbox.

A file search tool stored in a toolbox.

The FixedRatioVersionSelectionRule.

The inputs for generating a voice agent. Only kind and name are always required. The authoring service expands these inputs into a full, editable VoiceAgentDefinition, which is then created through POST /agents. The generated instructions and audio/voice settings are stored as separate fields on the resulting agent definition, so the caller can edit or override any of them afterward via standard agent versioning.

The GitHub Copilot managed harness for prompt agents.

Header-based secret authentication for a telemetry endpoint. The resolved secret value is injected as an HTTP header.

The hosted agent definition.

Configuration specific to the invocations protocol.

Configuration specific to the WebSocket-based invocations protocol.

The ManagedAgentIdentityBlueprintReference.

Configuration specific to the MCP protocol.

An MCP tool stored in a toolbox.

Extension methods on OpenAI.Responses.McpTool that expose Foundry-project-specific configuration (such as the project connection identifier).

A tool for integrating memories into the agent.

Memory search options.

The input definition information for a Microsoft Fabric tool as used to configure an agent.

Security details for OpenApi anonymous authentication.

authentication details for OpenApiFunctionDefinition Please note this is the abstract base class. The derived classes available for instantiation are: OpenAPIAnonymousAuthenticationDetails, OpenApiProjectConnectionAuthenticationDetails, and OpenAPIManagedAuthenticationDetails.

The input definition information for an openapi function.

The OpenAPIFunctionEntry.

Security details for OpenApi managed_identity authentication.

Security scheme for OpenApi managed_identity authentication.

Security details for OpenApi project connection authentication.

Security scheme for OpenApi managed_identity authentication.

The input definition information for an OpenAPI tool as used to configure an agent.

An OpenAPI tool stored in a toolbox.

Identifies the registered Foundry agent to optimize (request-only). Skills, tools, and system_prompt are specified in options.optimization_config.

An OTLP (OpenTelemetry Protocol) telemetry export endpoint.

The PatchAgentOptions.

An agent session providing a long-lived compute sandbox for hosted agent invocations.

The ProjectAgentSkills sub-client.

The ProjectsAgentDefinition. Please note this is the abstract base class. The derived classes available for instantiation are: DeclarativeAgentDefinition, ExternalAgentDefinition, HostedAgentDefinition, VoiceAgentDefinition, and WorkflowAgentDefinition.

The ProjectsAgentRecord.

A factory class for creating instances of the models for mocking.

A tool that can be used to generate a response. Please note this is the abstract base class. The derived classes available for instantiation are: Azure.AI.Projects.Agents.FabricIQPreviewTool, BingGroundingTool, MicrosoftFabricPreviewTool, SharepointPreviewTool, AzureAISearchTool, OpenAPITool, BingCustomSearchPreviewTool, BrowserAutomationPreviewTool, AzureFunctionTool, CaptureStructuredOutputsTool, A2APreviewTool, Azure.AI.Projects.Agents.A2ATool, Azure.AI.Projects.Agents.WorkIQPreviewTool, Azure.AI.Projects.Agents.WebIQPreviewTool, MemorySearchPreviewTool, OpenAI.ProgrammaticToolCallingParam, and Azure.AI.Projects.Agents.ToolSearchTool.

The ProjectsAgentVersion.

The ProjectsAgentVersionCreationOptions.

A web search configuration for bing custom search.

Promotion metadata recorded when a candidate is deployed to a Foundry agent.

Per-protocol configuration for the agent endpoint.

A record mapping for a single protocol and its version.

A PSTN destination for a telephony transfer target.

Declares where request/response text lives so content-safety guardrails can extract it.

An SSE event-type to text-field selector for streaming invocation output.

A reminder tool stored in a toolbox.

Configuration specific to the responses protocol.

Extension methods that convert OpenAI OpenAI.Responses.ResponseTool instances into the equivalent ProjectsAgentTool representation.

Session defaults applied to sessions created for a hosted agent version.

A single entry in a directory listing.

Response from uploading a file to a session sandbox.

A single Server-Sent Event frame emitted by the hosted agent session log stream. Each frame contains an event field identifying the event type and a data field carrying the payload as plain text. Although the current data payload is JSON-formatted, its schema is not contractual — additional keys may appear and the format may change over time. Clients should treat data as an opaque string and optionally attempt JSON parsing. New event types may be added in the future. Clients should gracefully ignore unrecognized event types. Wire format:

event: log
data: {"timestamp":"2026-03-10T09:33:17.121Z","stream":"stdout","message":"Starting server on port 18080"}
event: log
data: {"timestamp":"2026-03-10T09:34:52.714Z","stream":"status","message":"Successfully connected to container"}

The sharepoint grounding tool parameters.

The input definition information for a sharepoint tool as used to configure an agent.

A shell tool stored in a toolbox. This model is additive to toolbox configuration and does not modify the OpenAI tool contract or existing toolbox tool definitions.

A SIP destination for a telephony transfer target.

A deleted skill.

Inline content for defining a simple skill without uploading files. Follows the agentskills.io SKILL.md specification.

A reference to a versioned Foundry skill.

A specific version of a skill.

A deleted skill version.

An structured input that can participate in prompt template substitutions and tool argument binding.

A structured output that can be produced by the agent.

A Microsoft Teams Phone Extension binding owned by a voice agent.

A Microsoft Teams Phone Extension binding returned in a list, including its entity tag.

A Microsoft Teams destination for a telephony transfer target.

Customer-supplied telemetry configuration for exporting container logs, traces, and metrics.

A telemetry export endpoint configuration. Please note this is the abstract base class. The derived classes available for instantiation are: OtlpTelemetryEndpoint.

Authentication configuration for a telemetry endpoint. Please note this is the abstract base class. The derived classes available for instantiation are: HeaderTelemetryEndpointAuth.

A telephony binding owned by a voice agent. Please note this is the abstract base class. The derived classes available for instantiation are: TeamsPhoneExtensionTelephonyBinding and TwilioTelephonyBinding.

A telephony binding returned in a list, including its entity tag. Please note this is the abstract base class. The derived classes available for instantiation are: TeamsPhoneExtensionTelephonyBindingListItem and TwilioTelephonyBindingListItem.

A durable direct or campaign-created outbound call intent.

A cancellation request recorded for an outbound call job.

The optional execution window for a direct outbound call.

A bounded durable observation in the lifecycle of one telephony call.

Detailed diagnostics for a durable inbound call to a voice agent.

A summary of a durable inbound call to a voice agent.

Detailed provider-neutral timing for an inbound telephony call.

Correlation from a durable telephony call record to its customer-facing Foundry trace.

The destination of an outbound call.

The frozen fixed-interval retry policy returned for an outbound call or campaign.

The retry policy for one durable outbound call intent. max_attempts includes the first attempt. Strategy-specific settings are defined by the derived policy. Please note this is the abstract base class. The derived classes available for instantiation are: TelephonyOutboundFixedIntervalRetryPolicyResult.

A destination for a telephony transfer target. Please note this is the abstract base class. The derived classes available for instantiation are: PSTNTelephonyTransferDestination, SipTelephonyTransferDestination, and TeamsTelephonyTransferDestination.

A named destination to which the voice agent may transfer a call.

The telephony transfer targets configured for one voice agent.

Policy configuration for a toolbox, including content safety and other governance settings.

A toolbox that stores reusable tool definitions for agents.

A toolbox search tool stored in a toolbox.

An automatically provisioned container environment for a shell tool stored in a toolbox.

An existing container environment for a shell tool stored in a toolbox.

An execution environment for a shell tool stored in a toolbox. This environment model is scoped to toolbox configuration and does not modify the OpenAI shell environment contract. Please note this is the abstract base class. The derived classes available for instantiation are: ToolboxShellContainerAutoEnvironment and ToolboxShellContainerReferenceEnvironment.

Network access policy for an automatically provisioned toolbox shell container. Please note this is the abstract base class. The derived classes available for instantiation are: ToolboxShellNetworkPolicyDisabled.

A network policy that disables outbound access from a toolbox shell container.

A skill source included in a toolbox. Please note this is the abstract base class. The derived classes available for instantiation are: ToolboxSkillReference.

A reference to an existing skill to include in a toolbox.

An abstract representation of a tool stored in a toolbox. Please note this is the abstract base class. The derived classes available for instantiation are: A2APreviewToolboxTool, A2AToolboxTool, AzureAISearchToolboxTool, BrowserAutomationPreviewToolboxTool, CodeInterpreterToolboxTool, FabricIQPreviewToolboxTool, FileSearchToolboxTool, MCPToolboxTool, OpenApiToolboxTool, ReminderPreviewToolboxTool, ShellToolboxTool, ToolSearchToolboxTool, ToolboxSearchPreviewToolboxTool, WebIQPreviewToolboxTool, WebSearchToolboxTool, and WorkIQPreviewToolboxTool.

A specific version of a toolbox.

The versions associated with a toolbox.

A project connection resource.

A toolbox search tool stored in a toolbox.

A Twilio binding owned by a voice agent.

A Twilio binding returned in a list, including its entity tag.

The UpdateToolboxRequest.

Version indicator determining which agent version backs the session. Please note this is the abstract base class. The derived classes available for instantiation are: VersionRefIndicator.

Version indicator that references a specific agent version by name.

The VersionSelectionRule. Please note this is the abstract base class. The derived classes available for instantiation are: FixedRatioVersionSelectionRule.

The VersionSelector.

The audio configuration for a voice agent. These values are session defaults and may be overridden when connecting.

Input audio configuration for a voice agent.

Output audio configuration for a voice agent. Provider-specific fields are selected by voice_type:

• `openai`: `voice` and `speed`.
• `azure-standard`: `voice`, `voice_locale`, `speed`, `voice_temperature`, `custom_lexicon_url`,

• `azure-custom`: all `azure-standard` fields except `style`, plus `custom_voice_endpoint_id`.
• `azure-personal`: all `azure-standard` fields except `style`, plus `personal_voice_model`.
• `avatar-voice-sync`: all `azure-standard` fields except `voice` and `style`, plus `personal_voice_model`; the voice name is derived from the avatar.
• `azure-realtime-native`: `voice` and `speed`.

Avatar configuration for a voice agent. These values are session defaults and may be overridden when connecting.

Avatar placement and motion settings.

The avatar video background.

The rectangular crop applied to avatar video.

Avatar video encoder and presentation settings.

The avatar video resolution.

English-optimized Azure semantic voice activity detection.

Multilingual Azure semantic voice activity detection.

Azure semantic voice activity detection.

The voice agent definition. Its configuration (model, instructions, audio, tools, and optional avatar) drives a managed speech-to-speech experience. Establish realtime voice sessions through GET /agents/{agent_name}/endpoint/protocols/voice. Every create or update produces a new immutable version.

Server-side echo cancellation settings for input audio.

A service-managed control that ends the active conversation.

Semantic end-of-utterance detection configuration.

A native function tool executed by the client.

Session-start greeting configuration for a voice agent. Please note this is the abstract base class. The derived classes available for instantiation are: VoiceAgentLlmGeneratedGreetingConfig and VoiceAgentTemplateGreetingConfig.

Asynchronous input-audio transcription configuration. Extends the OpenAI Realtime transcription options with the Azure and MAI transcription models, custom speech models, and phrase hints.

Fields shared by interim-response configurations. Please note this is the abstract base class. The derived classes available for instantiation are: VoiceAgentLlmInterimResponseConfig and VoiceAgentStaticInterimResponseConfig.

A greeting authored by the session model from a scoped opening-turn prompt.

An interim response generated by a language model.

An MCP tool available to a voice agent.

Input audio noise reduction configuration.

OpenAI semantic VAD turn-detection settings.

Server-side voice activity detection.

A static interim response selected from configured text.

A sibling Foundry text agent that a voice agent may consult as a background specialist.

Configuration for sibling Foundry text agents that a voice agent may consult.

Policy for delivering responses while a voice agent waits for a subagent.

A service-managed control that acts on the active voice session without customer code or external authentication.

A deterministic greeting rendered with the voice agent's structured inputs and synthesized without model-authored generation.

A tool usable by a voice agent. Please note this is the abstract base class. The derived classes available for instantiation are: VoiceAgentFunctionTool, VoiceAgentMcpTool, VoiceAgentSystemTool, and VoiceAgentToolboxTool.

A reference to a Foundry toolbox, which is a versioned bundle of tools executed through its MCP endpoint.

Turn-detection configuration for a voice agent. Please note this is the abstract base class. The derived classes available for instantiation are: VoiceAgentAzureSemanticVadEnTurnDetection, VoiceAgentAzureSemanticVadMultilingualTurnDetection, VoiceAgentAzureSemanticVadTurnDetection, VoiceAgentSemanticVadTurnDetection, and VoiceAgentServerVadTurnDetection.

Metadata for a single conversation item's audio segment. For bring-your-own-storage (BYOS), the response includes blob_uri, a direct customer-storage URI without a SAS token, that the customer accesses with their own credentials. For Foundry-managed storage, blob_uri is absent and the bytes are streamed through the item's /audio/content route.

A persisted voice conversation. The Foundry envelope that owns a voice agent's stored transcript, responses, per-turn metrics, and audio. It is the parent, retention, and delete boundary: deleting it cascades to its responses, items, metrics, and audio. When finalization fails, any partial persisted responses, items, and item audio remain readable.

An engine that owns conversation handling for a voice agent. Please note this is the abstract base class. The derived classes available for instantiation are: VoiceHostedAgentConversationEngine.

Metadata for a conversation item's generated audio. For bring-your-own-storage (BYOS), the response includes blob_uri, a direct customer-storage URI without a SAS token, that the customer accesses with their own credentials. For Foundry-managed storage, blob_uri is absent and the bytes are streamed through the item's /audio/generated/content route.

A closed reference to the hosted text agent that owns conversation handling for a voice agent. The hosted agent is resolved within the same project and must support the invocations_ws protocol, Voice Live compatibility, and Bridge Protocol 1.0.

Metadata for the merged, whole-call stereo recording of a voice conversation (user audio on the left channel, agent audio on the right). Built once from the per-turn segments after the session ends and durably cached. The common metadata (format, sample rate, channels, channel layout, duration) is returned for both Foundry-managed and bring-your-own-storage (BYOS) recordings. For BYOS the response also includes blob_uri, the URI of the recording in the customer's own storage (no SAS token), which the customer downloads using their own storage credentials. For Foundry-managed storage blob_uri is absent and the bytes are streamed via the /audio/content route instead.

The role assigned to each channel of a merged stereo voice recording.

A persisted voice response representing one model inference turn within a conversation. In list results the output projection may be omitted; retrieve the full response (GET .../responses/{response_id}) or the paged response-items route (GET .../responses/{response_id}/items) for its output items. created_at/completed_at are Foundry durable ordering extensions.

Audio configuration for a response. Follows the OpenAI Realtime GA audio object shape.

The flat response audio-output projection, with optional voice, voice_type, voice_locale, and format fields.

Properties shared by persisted voice responses.

A WebIQ tool stored in a toolbox.

A web search tool stored in a toolbox.

Extension methods that expose Microsoft Foundry-specific configuration on OpenAI.Responses.WebSearchTool instances.

The workflow agent definition. Microsoft Foundry is retiring workflows on December 1, 2026. If you're looking to build new workflows, use Microsoft Agent Framework. To migrate existing workflows, see the Migration guide.

A WorkIQ tool stored in a toolbox.

Supported A2A protocol versions.

An access boundary for the activity protocol.

The status of an agent identity, applicable to both the agent instance identity and the agent blueprint.

The status of an agent session.

The level at which evaluation is performed.

Extensible status values shared by Foundry jobs.

The operational state of an agent.

Indicates the source of an agent's operational state. Empty when the state is not derived from a specific source.

The provisioning status of an agent version.

Available query types for Azure AI Search tool.

How package dependencies are resolved at deployment time for a code-based hosted agent.

The type of digital worker.

The type of telemetry data to export.

The Microsoft Agent Certification review status of the Microsoft 365 store title published for an agent.

How an invocations request/response body is parsed to locate text for content-safety moderation.

Author-declared response shape for the invocations protocol.

Known SSE event types emitted by the hosted agent session log stream. Additional event types may be introduced in future versions.

The transport protocol for telemetry export.

The lifecycle status of a telephony binding.

The timestamp used as the basis for call duration.

Known service-generated reasons that one telephony call ended, rather than reasons for an overall outbound call job. Additional string codes may be returned.

The lifecycle status of a durable outbound call job.

Known terminal reasons for an overall outbound call job, which can span multiple provider attempts. These are distinct from individual call lifecycle reasons. Additional string codes may be returned.

A provider-neutral lifecycle event name. Known values are stable; additional values may be added over time.

The outcome of one telephony lifecycle observation.

Known service-generated reasons for a telephony lifecycle event. An event reason does not necessarily describe the final outcome of the call. Additional string codes may be returned.

The component that supplied a telephony lifecycle observation.

The provider-neutral phase reached by an inbound telephony call.

The lifecycle status of an inbound telephony call.

The source of a telephony lifecycle timestamp.

The mode used to expose a telephony call as a customer-facing Foundry trace.

The availability status of a customer-facing telephony call trace.

The type of destination for an outbound call.

A telephony provider supported by an agent binding. Known values are stable; additional values may be added over time.

The transport used to deliver the avatar video stream.

The avatar type.

The semantic end-of-utterance detection model.

The sensitivity threshold for semantic end-of-utterance detection.

The input-audio transcription model identifier. This is a model name, not a Foundry deployment name. Mirrors the transcription models supported by the managed voice backend, covering the OpenAI Realtime transcription models plus the Azure and MAI models. Additional values may be added over time.

A condition that may trigger an interim response.

The input audio noise reduction mode.

A service-managed voice-session control action. Known values are stable; additional values may be added over time.

When a tool invocation creates a follow-up response. Additional values may be added over time.

An audio codec. Additional values may be added over time.

An audio container format. Additional values may be added over time.

A voice-audio participant role. Additional values may be added over time.

The lifecycle status of a persisted voice conversation:

• `in_progress`: the live session is active, or post-session persistence finalization is pending.
• `completed`: finalization succeeded after normal or client close, `end_conversation`, a max-duration `1001`

• `failed`: a terminal service, bridge, storage, or unrecoverable transport failure prevented finalization.

How the model backing a voice agent is served. This is independent of the architecture (realtime or cascaded), which the service derives from the selected model.

An output modality the agent may produce. animation and avatar are used when an avatar is configured.

The voice implementation. Additional values may be added over time.

The version of the service to use.

Feature opt-in keys for agent definition operations supporting conditional preview features.

An output-audio timestamp kind supported by a voice agent.

The source of reference audio used for echo cancellation.

Additional fields that a voice-agent session may include in service outputs.