[fern-api]

⚠️ 2.0.0 → 3.0.0

Breaking: Multiple public classes and modules are renamed (e.g., TrieveCredential → MicrosoftCredential, User → UserWithRole, GladiaTranscriberLanguages → GladiaTranscriberLanguagesItem), and two enum modules are restructured into Union classes (InviteUserDtoRole, UpdateUserRoleDtoRole), all of which will cause NameError or NoMethodError for existing callers without code changes.

Breaking Changes

• UpdateAssistantDtoCredentialsItem — the TRIEVE credential variant (CreateTrieveCredentialDto) has been removed from the union. Update any code passing a Trieve credential to use an alternative credential type.

Added

• UpdateAssistantDtoCredentialsItem — two new credential variants: CreateS3CompatibleCredentialDto (key S_3_COMPATIBLE) and CreateMicrosoftCredentialDto (key MICROSOFT).
• CreateCallDto — new optional fields assistant_version, squad_version, and a typed transport field (replacing the previous untyped hash).
• sort_by parameter — added to list/paginated endpoints across Chats, Eval, Insight, ObservabilityScorecard, PhoneNumbers, ProviderResources, and Sessions clients for finer-grained result ordering.
• New filter parameters — id_any on Chats and Sessions list; squad_overrides on Sessions list; search on eval runs pagination; purpose on Files list and upload; metadata on file upload.
• Vapi::Board::Client — new reporting client with full CRUD support for boards (board_controller_find_all, board_controller_create, board_controller_find_one, board_controller_remove, board_controller_update) and a board_controller_metrics_overview_ensure method for the default metrics-overview board.
• Call artifact download methods — seven new methods on Vapi::Calls::Client for downloading call recordings and logs: call_artifact_controller_mono_recording_download, stereo_recording_download, video_recording_download, customer_recording_download, assistant_recording_download, pcap_download, and call_logs_download.
• New transcriber and voice providers — XaiTranscriber and VapiTranscriber added to UpdateAssistantDtoTranscriber; XaiVoice and MicrosoftVoice added to UpdateAssistantDtoVoice.
• New enum values — CALL_ARTIFACT_UPLOAD added to UpdateAssistantDtoServerMessagesItem; CANCELLED and ARCHIVED added to CampaignControllerFindAllRequestStatus; sort_by field and CampaignControllerFindAllRequestSortBy enum added to campaign list requests.
• Vapi::KnowledgeBasesV2::Client — new client for managing v2 knowledge bases, including CRUD operations and file attachment, detachment, and retry methods.
• Vapi::Simulations::Client and related clients (SimulationPersonalities, SimulationScenarios, SimulationRuns, SimulationSuites) — new clients for simulation workflows.
• Vapi::TrafficAllocations::Client and Vapi::Board::Client — new top-level namespace clients accessible directly from the root Vapi::Client.
• sort_by optional field added to ListChatsRequest, EvalControllerGetPaginatedRequest, and InsightControllerFindAllRequest; id_any filter field added to ListChatsRequest.
• ListFilesRequest and CreateFilesRequestPurpose — new types for filtering files by purpose (assistant, composer-attachment, knowledge-base-v2); assistant_id optional field added to InsightRunDto.
• Vapi::SimulationPersonalities::Client — new client for managing simulation personalities, with methods to list, create, retrieve, update, and delete personalities used in AI tester simulations.
• sort_by field — new optional sort_by parameter (values: createdAt, duration, cost) added to ScorecardControllerGetPaginatedRequest, PhoneNumberControllerFindAllPaginatedRequest, ProviderResourceControllerGetProviderResourcesPaginatedRequest, and ListSessionsRequest for finer-grained result ordering.
• ListSessionsRequest#squad_overrides — new optional field for filtering sessions by squad overrides.
• ListSessionsRequest#id_any — new optional field for filtering sessions by any matching ID.
• Vapi::SimulationScenarios::Client — new client for managing simulation scenarios, supporting list, create, get, update, and delete operations against the eval/simulation/scenario endpoint.
• Vapi::SimulationSuites::Client — new client for managing simulation suites, supporting list, create, duplicate, get, update, and delete operations against the eval/simulation/suite endpoint.
• Vapi::SimulationSuites::Types::CreateSimulationSuiteDto and UpdateSimulationSuiteDto — request DTOs for creating and updating simulation suites, including fields for name, slack_webhook_url, simulation_ids, target_assignments, and path.
• Vapi::SimulationScenarios::Types::UpdateScenarioDto — request DTO for updating scenarios, including instructions, evaluations, hooks, target_overrides, tool_mocks, and latency_expectations.
• New enum types ScenarioControllerFindAllRequestSortBy, ScenarioControllerFindAllRequestSortOrder, SimulationSuiteControllerFindAllRequestSortBy, and SimulationSuiteControllerFindAllRequestSortOrder for filtering and sorting list results.
• Vapi::Simulations::Client — new client for the Simulations API, supporting scenario generation (simulation_generate_controller_generate), listing (simulation_controller_find_all), creating (simulation_controller_create), fetching (simulation_controller_find_one), deleting (simulation_controller_remove), updating (simulation_controller_update), and concurrency checks (simulation_controller_get_concurrency).
• KnowledgeBaseTool and GhlTool — added as new union members to all tool response types (GetToolsResponse, ListToolsResponseItem, CreateToolsResponse, DeleteToolsResponse, UpdateToolsResponse).
• CreateCodeToolDto and UpdateCodeToolDto — added as new CODE union members to CreateToolsRequest and UpdateToolsRequestBody respectively.
• StructuredOutputControllerRunResponse, StructuredOutputControllerFindAllRequestSortBy, and UpdateStructuredOutputDtoConditionsItem — new types supporting structured output run results and filtering.
• id_any filter field on ListSquadsRequest and sort_by field on StructuredOutputControllerFindAllRequest — new optional query parameters for finer-grained list filtering.
• Vapi::TrafficAllocations::Client — new client for the beta traffic-splitting API, supporting paginated history, creating allocations, fetching the latest allocation, and retrieving a single allocation by ID.
• tool_refs and reasoning_effort fields — added as optional fields on AnthropicModel, AnthropicBedrockModel, and AnyscaleModel to support tool references and reasoning-effort control; AnthropicBedrockModel also gains fallback_models.
• AssemblyAiTranscriber enhancements — new optional fields mode, prompt, agent_context, agent_context_auto_update_enabled, and language_codes; new AssemblyAiTranscriberMode (max_accuracy, min_latency, balanced) and AssemblyAiTranscriberLanguageCodesItem enums; new speech models universal-3-5-pro and universal-3-6-pro.
• Assistant#latest_version and Assistant#model_deprecations — new optional fields exposing the assistant's latest published version string and any model deprecation notices.
• New enum values and type fields — AnthropicModelModel gains claude-sonnet-5 and claude-haiku-5-5; AnthropicBedrockCredentialRegion gains eu-central-1; AnalysisCost gains structured_output_breakdown; AssistantActivation gains assistant_version and squad_version.
• AssistantDraft — new type representing a versioned draft of an assistant, with full paginated response support via AssistantDraftPaginatedResponse and AssistantDraftPaginatedMetadata.
• Conflict response types — AssistantDraftConflictResponseDto, AssistantManagedConflictResponseDto, and AssistantPinnedConflictResponseDto for handling draft conflict scenarios.
• VapiModel — added as a new provider member to AssistantModel and AssistantOverridesModel unions.
• XaiTranscriber and VapiTranscriber — added as new provider members to AssistantTranscriber and AssistantOverridesTranscriber unions; XaiVoice and MicrosoftVoice added to AssistantOverridesVoice.
• CALL_ARTIFACT_UPLOAD — new enum value added to AssistantServerMessagesItem and AssistantOverridesServerMessagesItem.
• AssistantVersion and supporting types (AssistantVersionTranscriber, AssistantVersionModel, AssistantVersionVoice, AssistantVersionVoicemailDetection, AssistantVersionPaginatedMetadata, and related enums) for managing versioned assistant configurations.
• Board, BoardInsightItem, BoardItemPosition, and BoardItemSize types for dashboard board support, including layout and time-range override fields.
• AudioFormat and BackgroundSoundUrlValidationResult types, along with their associated enum modules (AudioFormatFormat, BackgroundSoundUrlValidationResultReason).
• New enum values across several types: GPT-5.6 and GPT-4.1 model variants in AzureOpenAiCredentialModelsItem; switzerlandnorth and switzerlandwest in AzureCredentialRegion and AzureOpenAiCredentialRegion; XAI and MICROSOFT voice providers in AssistantVoice.
• New optional fields latest_version and baseline_version on BashTool, and system_key on BarInsight and Board.
• CampaignContact, CampaignContactWithOutcome, CampaignContactCounters, CampaignContactPaginatedResponse, CampaignSummary, CampaignCallMetrics, CampaignPredialPlan, and CampaignServerMessagesItem — new types for managing campaign contacts and tracking campaign progress.
• BoardItemsItem, BoardLayout, BoardMetricWidgetItem, BoardMetricWidgetItemType, and BoardPaginatedResponse — new types for analytics board and metric widget support.
• BooleanComparatorScorecardMetricCondition and related enum types — new types for defining boolean scorecard metric conditions.
• CallTransport — new union type discriminated by provider, supporting VAPI_WEBSOCKET, VONAGE, TWILIO, VAPI_SIP, TELNYX, and DAILY transport variants.
• CallEndedReason — extended with new values for xAI/Microsoft voice and transcriber errors, Cartesia transcriber failures, SIP outbound scenarios, ElevenLabs concurrent-request and voice-disabled errors, squad/version validation errors, and more; CampaignStatus gains CANCELLED and ARCHIVED values; BotMessage gains optional assistant_name and assistant_id fields; Campaign gains optional max_concurrency, assistant_overrides, squad_overrides, server, server_messages, and predial_plan fields.
• CampaignSummaryPaginatedResponse, CampaignSummaryServerMessagesItem, and CampaignSummaryStatus — new types supporting campaign management with paginated results and status/event enumerations.
• ClientInboundMessageAppendContext — new inbound message type (with kind and content fields) registered as the APPEND_CONTEXT variant in the ClientInboundMessageMessage union.
• assistant_version — new optional field added to all ClientMessage* types; ClientMessageTranscript also gains assistant_id, assistant_name, confidence, and confidence_source fields.
• New Cartesia capabilities — CartesiaVoiceModel gains sonic-3.5 and sonic-3.6 model variants; CartesiaTranscriberModel gains ink-2; CartesiaVoiceLanguage gains Odia (or) and Urdu (ur); CartesiaCredential gains an optional api_url field.
• tool_refs on CerebrasModel and latest_version/baseline_version on CodeTool — new optional fields for tool reference linking and version tracking.
• CreateAssistantDraftDto — new type (and full suite of supporting union/enum types) for creating assistant drafts with configurable transcriber, model, voice, and behavior fields.
• XaiTranscriber and VapiTranscriber — new transcriber options available in CreateAssistantDtoTranscriber and ConversationNodeTranscriber.
• XaiVoice and MicrosoftVoice — new voice provider options available in CreateAssistantDtoVoice and ConversationNodeVoice.
• ConflictErrorBody — new union type discriminating between TOOL_PINNED and TOOL_WRITE_CONFLICT conflict error responses.
• New region and field additions — eu-central-1 added to CreateAnthropicBedrockCredentialDtoRegion; switzerlandnorth/switzerlandwest added to CreateAzureCredentialDtoRegion; latestVersion and baselineVersion optional fields added to ComputerTool; VapiModel added to CreateAssistantDtoModel.
• CreateCampaignDto — new type for configuring outbound calling campaigns with assistant, squad, or workflow selection, customer lists, dial plans, and scheduling.
• CreateCampaignDtoServerMessagesItem — new enum covering campaign and contact lifecycle server-message events (e.g. campaign.started, contact.dispatched).
• CreateElevenLabsCredentialDtoApiUrl — new enum for selecting the ElevenLabs API endpoint, including EU residency; exposed as optional api_url on CreateElevenLabsCredentialDto.
• New enum values added to CreateAzureOpenAiCredentialDtoModelsItem (GPT-5.6 Luna/Terra/Sol, GPT-4o, GPT-4.1, GPT-5.4-mini-2026-03-17) and CreateAzureOpenAiCredentialDtoRegion (switzerlandnorth, switzerlandwest).
• Optional fields api_url on CreateCartesiaCredentialDto and squad_overrides on CreateCustomerDto for finer-grained credential and customer configuration.
• CreateSimulationRunResponse — new type representing the response from creating a simulation run, including status, timing, item counts, target, and transport fields.
• CreateToolDraftDto — new type for creating draft tools, with supporting enum types CreateToolDraftDtoType, CreateToolDraftDtoMethod, and CreateToolDraftDtoVerb.
• CreateS3CompatibleCredentialDto — new credential type for configuring S3-compatible storage buckets.
• CreateOutboundCallDtoTransport and CreateTrafficAllocationTargetDto — new types for outbound call transport selection and assistant version traffic allocation.
• CreateStructuredOutputDto#conditions, CreateScenarioDto#latency_expectations, CreateSonioxCredentialDto#api_url, and CreateWebCallDto#assistant_version/squad_version — new optional fields on existing DTOs.
• XaiVoice and MicrosoftVoice added as voice options in CreateWorkflowDtoVoice; XaiTranscriber and VapiTranscriber added to CreateWorkflowDtoTranscriber.
• latestVersion and baselineVersion optional fields added to DtmfTool and EndCallTool; toolRefs optional field added to CustomLlmModel, DeepInfraModel, and DeepSeekModel.
• ElevenLabsCredentialApiUrl enum and api_url optional field added to ElevenLabsCredential for EU residency endpoint support; speed and expressivity optional fields added to DeepgramVoice; 43 new Deepgram voice IDs added to DeepgramVoiceId.
• New enum values added across DeepSeekModelModel (deepseek-flash, deepseek-flash-thinking), EvalAnthropicModelModel (claude-sonnet-5, claude-haiku-5-5), EvalGoogleModelModel (Gemini 3.x variants), EvalOpenAiModelModel (GPT-5.x and regional variants), and DeepgramVoiceModel (flux).
• New types CustomerSpeechTimeoutOptionsTriggerResetMode, DeepgramTranscriberRedactionItem, EndedReasonCondition, EndedReasonConditionOperator, and ExportChatDtoSortBy added; path field added to EvaluationPlanItem; sortBy and idAny fields added to ExportChatDto.
• FallbackMicrosoftVoice — new fallback voice provider for Microsoft Azure TTS, supporting voice ID, style, style degree, role, speed, and chunk plan configuration; registered as a MICROSOFT member in FallbackPlanVoicesItem.
• FallbackXaiVoice — new XAI fallback voice option added to FallbackPlanVoicesItem.
• FallbackAssemblyAiTranscriber — new optional fields mode, prompt, agentContext, agentContextAutoUpdateEnabled, and languageCodes added, along with new FallbackAssemblyAiTranscriberMode and FallbackAssemblyAiTranscriberLanguageCodesItem enums.
• FallbackDeepgramVoice — new optional speed and expressivity fields; FallbackDeepgramVoiceId expanded with 43 new voice IDs; FallbackDeepgramVoiceModel gains flux; FallbackDeepgramTranscriberRedactionItem enum added.
• New enum values and fields across ExportSessionDto (idAny, sortBy), FallbackSonioxTranscriber (languages, endpointSensitivity, endpointLatencyAdjustmentLevel, contextGeneral, confidenceThreshold), Cartesia voice models (sonic-3.5, sonic-3.6 series), Rime AI voice IDs and models (coda, mistv3), Google transcriber models (gemini-3.x series), ElevenLabs eleven_v4_turbo, and OpenAI voice IDs.
• tool_refs — new optional field on GoogleModel, GroqModel, and InflectionAiModel for referencing tools by ToolRef objects.
• KnowledgeBaseTool, KnowledgeBaseV2, KnowledgeBaseV2File, KnowledgeBaseV2WithFiles — new types for managing knowledge bases and their associated files via the v2 API.
• LatencyEvaluationResult — new type (with LatencyEvaluationResultMetric and LatencyEvaluationResultAggregation enums) for representing latency evaluation outcomes.
• New Gemini model enum values (GEMINI_38_FLASH, GEMINI_35_FLASH, GEMINI_31_FLASH_LITE) added to GoogleModelModel, GoogleTranscriberModel, and KnowledgeBaseModel.
• New event constants added to JsonQueryOnEventsTableOn, including SERVER_MESSAGE_REQUEST_SUCCEEDED, SERVER_MESSAGE_REQUEST_FAILED, ASSISTANT_KNOWLEDGE_BASE_REQUEST_SUCCEEDED, ASSISTANT_MODEL_RESPONSE_EMPTY, and analysis/pipeline events; ContextEngineeringPlanPreviousAssistantMessages added to handoff destination context engineering plan unions.
• MicrosoftVoice — new voice provider type for Microsoft Azure TTS, with support for MicrosoftVoiceVoiceId, MicrosoftVoiceStyle, MicrosoftVoiceRole, and MicrosoftCredentialProvider.
• LatencyExpectation, LatencyExpectationMetric, and LatencyExpectationAggregation — new types for defining and evaluating call latency thresholds by metric and aggregation method.
• ModelDeprecationNotice and ModelDeprecationNoticeReplacementStatus — new types for surfacing model deprecation and retirement information.
• LegacyAssistantVersion and LegacyAssistantVersionPaginatedResponse — new types for retrieving paginated assistant version history.
• New fields on existing types — OpenAiModel gains tool_refs, speaker, reasoner, service_tier, and reasoning_effort; ModelCost gains seconds, usage_complete, reasoning_tokens, and service_tier_breakdown; MakeTool and McpTool gain latest_version and baseline_version; LineInsight gains system_key; new GPT-5.x/6 variants added to OpenAiModelFallbackModelsItem.
• S3CompatibleStorageCredential, S3CompatibleBucketPlan, and S3CompatibleStorageCredentialProvider — new types enabling S3-compatible storage credential configuration.
• ServerMessageCampaignPredial and ServerMessageCallArtifactUpload — new server message types for campaign pre-dial events and call artifact upload notifications.
• assistant_version — new optional field added to all server message types (e.g. ServerMessageAssistantRequest, ServerMessageEndOfCallReport, ServerMessageConversationUpdate, and others) to surface the assistant version on each event.
• New RimeAiVoiceIdEnum values (CLEMENTINE, WALNUT, EYRE, BANCROFT, HESSE, BEATTY, GODFREY) and new RimeAiVoiceModel values (CODA, MISTV_3) expanding Rime AI voice options.
• New types and fields: RevokeInvitationResponseDto, ScenarioInUseConflictResponseDto, SayHookActionExact, ScorecardMetricConditionsItem, latency_expectations on Scenario, and many new ServerMessageEndOfCallReportEndedReason enum values covering xAI, Microsoft, Cartesia, and SIP error scenarios.
• assistant_version — new optional field added to all ServerMessage* types, exposing the assistant version on every server-sent event.
• ServerMessageCampaignPredial and ServerMessageCallArtifactUpload — two new server message variants added to the ServerMessageMessage union; ServerMessageResponseCampaignPredial added to ServerMessageResponseMessageResponse.
• ServerMessageTranscript — new optional fields assistant_id, assistant_name, confidence, and confidence_source (typed as ServerMessageTranscriptConfidenceSource) for richer transcript metadata.
• SimulationRunListItem, SimulationRunListSummary, SimulationRunListSource, SimulationRunPaymentRequiredResponse, and related enums — new types supporting simulation run listing and payment-required responses.
• ServerMessageStatusUpdateEndedReason — new constants for xAI, Microsoft, Cartesia, VAPI transcriber, SIP outbound, and squad/version validation failure scenarios.
• ToolVersion, ToolDraft, and ToolRef — new types supporting tool versioning and draft management workflows, including paginated list responses (ToolVersionPaginatedResponse, ToolDraftPaginatedResponse).
• TrafficAllocation — new type for managing assistant traffic allocation, with supporting enums TrafficAllocationActorType and TrafficAllocationAllocationIntent, and TrafficAllocationLatestResponseDto.
• TelnyxTransport — new type exposing Telnyx call transport metadata (conversationType, callControlId, callLegId, callSessionId).
• ToolCallResultMessageWarning and ToolCallResultSpokenMessage — new types for tool call result warnings and spoken message variants; ToolCallResultMessage gains an optional warnings field.
• New optional fields on existing types: tool_refs on TogetherAiModel, role on ToolMessageFailed, latest_version/baseline_version on TextEditorTool, and system_key on TextInsight.
• TrafficAllocation* types — new TrafficAllocationPaginatedResponse, TrafficAllocationTarget, TrafficAllocationStaleConflictResponseDto, TrafficAllocationVersionConflictResponseDto, and related enum for managing traffic allocation resources.
• TransferArtifact and supporting types — TransferArtifactDestination, TransferArtifactMode, TransferArtifactStatus, and TransferArtifactMessagesItem expose call transfer metadata including destination, mode, transcript, and status.
• TransientTwilioPhoneNumber — new type (with TransientTwilioPhoneNumberFallbackDestination and TransientTwilioPhoneNumberHooksItem) for transient Twilio phone number configuration.
• UpdateAssistantDraftDto — new DTO and its full set of supporting union and enum types for updating assistant drafts.
• TwilioTransport / TranscriptWordConfidence — new TwilioTransport type for Twilio transport details and TranscriptWordConfidence for per-word transcription confidence data; optional name field added to TransferDestinationAssistant, TransferDestinationNumber, and TransferDestinationSip; XAI and VAPI transcriber variants and XAI and MICROSOFT voice variants added to transfer assistant unions.
• UpdateAssistantDraftDto supporting types — new union and enum types for credentials, first-message mode, hooks, model, server messages, transcriber, voice, and voicemail detection to support assistant draft updates.
• UpdateCampaignDto and UpdateCampaignDtoStatus — new types for updating outbound calling campaigns, including name, assistant/workflow/squad IDs, phone number, dial plan, schedule, and status.
• UpdateAssistantVersionMetadataDto — new type for setting versionName and versionDescription on assistant versions.
• provider field added as an optional field to all Update*CredentialDto types (Azure, AzureOpenAI, Cartesia, Cerebras, Cloudflare, CustomCredential, CustomLlm, DeepInfra, DeepSeek, Deepgram, Email, and others).
• New enum values added to UpdateAzureOpenAiCredentialDtoModelsItem (GPT-5.6 Luna/Terra/Sol, GPT-4o, GPT-4.1, GPT-5.4-mini) and new Switzerland region variants (switzerlandnorth, switzerlandwest) to Azure region enums.
• provider field added as an optional field to 20+ Update*CredentialDto types (e.g., UpdateGcpCredentialDto, UpdateGoogleCredentialDto, UpdateOpenAiCredentialDto, UpdateMistralCredentialDto, and others), each backed by a new single-value enum module.
• UpdateMicrosoftCredentialDto — new DTO for updating Microsoft credentials, with optional provider, api_key, region, and name fields.
• UpdateKnowledgeBaseToolDto — new DTO for updating knowledge-base tools, including optional messages, knowledge_base_id, function, and rejection_plan fields.
• type field added as an optional field to UpdateGhlToolDto, UpdateMakeToolDto, and UpdateOutputToolDto, each backed by a new enum module.
• Inline documentation comments added to several tool and insight DTO classes (e.g., UpdateFunctionToolDto, UpdateMcpToolDto, UpdateLineInsightFromCallTableDto).
• VapiTranscriber — new transcriber type with version, language, languages, keywords, and turn_taking fields; VAPI and XAI members added to WorkflowTranscriber and WorkflowUserEditableTranscriber.
• Transport types — VapiSipTransport, VapiWebCallTransport, VapiWebsocketTransport, and VonageTransport models added for SIP, web-call, WebSocket, and Vonage call transport metadata.
• Voice cloning support — VapiVoiceCloneDto, VapiVoiceLanguage, and VapiVoiceVersion types added; VoiceLibrary gains optional is_cloned and clone_backend fields; VoiceLibraryVoiceResponseAge union type added.
• Version-pinning conflict types — VersionPinConflictResponseDto, VersionPinReference, and VersionPinReferenceSourceType added for handling version-pin conflict responses.
• New model enum values and messages field — messages optional field added to all workflow model types; new GPT-5.x/6, Gemini 3.x, Claude Sonnet 5, and Claude Haiku 5.5 variants added to their respective model enums.
• XaiVoice — new voice type for xAI with support for voice ID selection, language, speed, chunk plan, and fallback plan.
• XaiTranscriber — new transcriber type for xAI with model, language, and fallback plan configuration.
• XaiVoiceVoiceId, XaiVoiceLanguage, XaiTranscriberModel, and XaiTranscriberLanguage — supporting enums for the new xAI voice and transcriber types.
• XaiModelModel — new model variants grok-4.20-0309-reasoning, grok-4.20-0309-non-reasoning, and grok-4.3 added to the xAI model enum.
• WorkflowVoice and WorkflowUserEditableVoice — new XAI and MICROSOFT union members added; XaiModel gains an optional tool_refs field.

---

See full changelog

---

⚠️ Version bump not verified

Automatic version analysis was unavailable or incomplete for this generation, so the version bump and changelog may understate the changes (a PATCH bump with no changelog is applied when no analysis is available). If this change is breaking or adds features, adjust the version and changelog before merging.

Reason: 2 of 36 FAI chunk requests failed; last error: TypeError: fetch failed