Repository navigation
feat: add claude-haiku-5-5, Sonic 3.6, Gemini 3.8-flash, and idempotent simulation runs - #50
fern-api[bot] wants to merge 5 commits into
Conversation
d3ba3dc to
b11338f
Compare
da4ebf9 to
c08cb43
Compare
8b38e63 to
aed7433
Compare
aff5143 to
08c602f
Compare
5ba990e to
8f98d61
Compare
ab02d99 to
211225c
Compare
211225c to
68663de
Compare
68663de to
6eea459
Compare
Generated by Fern CLI Version: unknown Generators: - fernapi/fern-typescript-sdk: 3.63.0
…flash, and idempotent simulation runs Expand model and voice coverage, introduce idempotency support for simulation runs, add new tool types and conflict response DTOs, and expose baseline versioning across all tool interfaces. Key changes: - Add `claude-haiku-5-5` to `AnthropicModelModel`, `AnthropicBedrockModelModel`, `AnthropicBedrockModelFallbackModelsItem`, and `EvalAnthropicModelModel`; introduce `AnthropicModelReasoningEffort` and `AnthropicBedrockModelReasoningEffort` enums for adaptive thinking control - Add `Sonic36` / `Sonic3620260827` to `CartesiaVoiceModel` and `FallbackCartesiaVoiceModel`; add Odia (`Or`) and Urdu (`Ur`) to `CartesiaVoiceLanguage` and `FallbackCartesiaVoiceLanguage` - Add `Gemini38Flash` to `GoogleModelModel`, `GoogleTranscriberModel`, `FallbackGoogleTranscriberModel`, `EvalGoogleModelModel`, and `KnowledgeBaseModel` - Add `Idempotency-Key` field to `CreateSimulationRunDto` and `reused` flag to `CreateSimulationRunResponse` for safe retries - Add `baselineVersion` optional field to all tool interfaces (`ApiRequestTool`, `BashTool`, `CodeTool`, `ComputerTool`, `DtmfTool`, `EndCallTool`, `FunctionTool`, `HandoffTool`, `KnowledgeBaseTool`, `MakeTool`, and GoHighLevel/Google Calendar tools) - Add `AssistantManagedConflictResponseDto`, `AssistantManagedConflictResponseDtoError`, and `AssistantManagedVapiAgentsDto` for Vapi Agents lock conflict responses - Add `ModelCostServiceTierBreakdown`, `UserWithRole`, `UserWithRoleRole`, `UserWithRoleRoleZero` exported types; remove `User`, `GhlToolType`, `UpdateCodeToolDtoType`, `CreateTrieveCredentialDto`, `CreateTrieveKnowledgeBaseDto`, `TrieveKnowledgeBase`, `TrieveKnowledgeBaseImport`, `UpdateTrieveKnowledgeBaseDto`, `AssistantPinnedConflictResponseDtoError`, `TrafficAllocationStaleConflictResponseDtoError` - Add `serverMessage.requestSucceeded`, `serverMessage.requestFailed`, and `pipeline.denoiserFailed` event values to `JsonQueryOnEventsTableOn` 🌿 Generated with Fern
6eea459 to
7fcf05c
Compare
SDK maintenance pilot — assessment of 76c8e83Disposition: not ready for release review. Existing CI and the focused installed-package checks below pass. Release/migration notes omit confirmed breaking changes, and live Vapito validation is incomplete. This comment is evidence for the named commit, not an approval. Human reviewer: Steve Smith (@stephenvapiai). Baseline and provenance
Findings and fix routing
Validation performed
Draft release/migration notesBreaking: remove the exported helpers/types listed above. Replace GHL/code discriminator constants with the documented literals, and use endpoint union variants when a discriminator is required. Rename direct references to the structured-output preview response type. Before: import { Vapi } from "@vapi-ai/server-sdk";
const ghl: Vapi.GhlToolType = Vapi.GhlToolType.Ghl;
const code: Vapi.UpdateCodeToolDtoType = Vapi.UpdateCodeToolDtoType.Code;
const preview: Vapi.StructuredOutputControllerRunResponseZero = { skipped: {} };After (compile-verified): import { Vapi } from "@vapi-ai/server-sdk";
const ghl: Vapi.GetToolsResponse.Ghl["type"] = "ghl";
const code: Vapi.UpdateToolsRequestBody.Code["type"] = "code";
const preview: Vapi.StructuredOutputControllerRunResponseOne = { skipped: {} };For removed conflict discriminator constants, narrow the response body and compare its literal Added: simulation-run idempotency request header and Changed: assistant-update and structured-output create/update/delete HTTP 409 responses now use typed Remaining validation and next actionLive Vapito validation: incomplete because this session has no configured Vapito connection/control-token access. No Vapito was created, no live requests were made, and no test resources need cleanup. Local results were obtained on macOS arm64 / Node 25.9.0; they do not establish the full supported-runtime matrix. Once access is available, run the installed candidate against a healthy seeded Vapito on a recorded backend SHA: assistant create/get/update/delete and a bounded changed-endpoint scenario. Record sandbox health, assertions, cleanup, and any backend/fixture limitations. Next: Steve to review intended removals and compatibility policy; incorporate accurate migration notes; complete Vapito validation. Reassess if the PR head changes. No repository code, release tag, or package publication was changed by this pilot. |
Breaking: MAJOR because the overall version bump was declared as such by the release pipeline; however, no specific symbol-level removals or breaking API changes were identified in the captured diff — maintainers should confirm whether unreported contract changes (such as removed exports or renamed types) justify the MAJOR designation.
Breaking Changes
Added
AnthropicModel.reasoningEffortandAnthropicBedrockModel.reasoningEffort— new optional fields to control adaptive thinking effort forclaude-haiku-5-5, accepting values fromnonetomax.baselineVersion— new optional display-only field added to all tool interfaces (McpTool,OutputTool,QueryTool,SipRequestTool,SlackSendMessageTool,SmsTool,TextEditorTool,TransferCallTool,VoicemailTool,GhlTool) returning the baseline label before a version is stored.ModelCostServiceTierBreakdown— new interface reporting per-OpenAI-service-tier cost and token counts, attached as the optionalserviceTierBreakdownarray onModelCost.UserWithRole,UserWithRoleRole, andUserWithRoleRoleZero— new exported types representing an org member with their role and optional last-active timestamp.{{messagesWithConfidence}}template variable — now supported in structured output Liquid templates across all workflow model types andStructuredOutput/CreateStructuredOutputDto.ClaudeHaiku55added toWorkflowAnthropicModelModelandWorkflowAnthropicBedrockModelModel;Gemini38Flashadded toWorkflowGoogleModelModel.Changed
OpenAiModelServiceTier— documentation updated to clarify thatpriorityis the legacy alias forfast,autois the effective default when unset, and Fast mode billing is per-model and only charged when OpenAI confirms the request was served in Fast mode.See full changelog