Vercel.AI

Client, text generation, streaming, tools, and agents.

Classes

Type	Summary
AbortPart	The call was aborted. It is the last part of the stream.
Agent	An agent that calls tools until StopWhen matches. Maps to Agent.
AgentOptions	Agent settings.
Ai	Static entry points. String model ids use the AI Gateway and AIGATEWAYAPIKEY.
AiClient	Default IAiClient. String model ids are resolved by the AI Gateway.
AiSdkOptions	Options for AiSdkOptions%7d).
AiSdkServiceCollectionExtensions	Dependency injection for IAiClient.
ApiOrProvider<TApi, TProvider>	An API or a provider that exposes one. Maps to the upstream Api \| ProviderV4 unions.
CustomProvider	A provider backed by caller-supplied models. Maps to customProvider. A model id that is not registered goes to FallbackProvider, or throws NoSuchModelError when there is none.
DefaultInstructionsMiddleware	Prepends system instructions when the prompt has none. Maps to defaultInstructionsMiddleware.
DefaultSettingsMiddleware	Fills sampling settings when the caller left them unset. Maps to defaultSettingsMiddleware.
EmbedEventValue<TOne, TMany>	A single item for embed or a list for embedMany, as carried by embed events.
EmbedOptions	Embedding options.
ErrorPart	The stream failed.
EvaluateOptions	Evaluation options.
ExecutedTool	The result of executing a tool.
ExtractJsonMiddleware	Strips markdown JSON fences. Maps to extractJsonMiddleware.
ExtractReasoningMiddleware	Splits delimited reasoning, such as …, out of text. Maps to extractReasoningMiddleware.
FinishPart	The full stream finished.
GenerateImageOptions	Image generation options.
GenerateSpeechOptions	Speech options.
GenerateTextOptions	Options for CancellationToken). Property names follow generateText.
GenerateTextResult	Result of generateText.
GenerateVideoOptions	Video options.
LanguageModelCallEndEvent	Passed to OnLanguageModelCallEnd after a provider model call returns.
LanguageModelMiddleware	Passes both operations through.
LanguageModelMiddlewareExtensions	Wraps a language model with middleware. Maps to wrapLanguageModel.
LanguageModelRequestMetadata	Metadata for a language model request. Maps to LanguageModelRequestMetadata.
LanguageModelResponseMetadata	Metadata for a language model response, as the finish-step part carries it. Maps to Omit<LanguageModelResponseMetadata, ‘messages’ \| ‘body’>.
OutputChunkTimingStats	Timing statistics for the gaps between generated output chunks, in milliseconds.
OutputSpec	Structured output selector. Maps to Output.
PrepareStepContext	Context passed to prepareStep.
PrepareStepUpdate	Optional changes applied before a step.
ProviderRegistry	Looks up provider models by provider:model ids. Maps to createProviderRegistry. Everything after the first separator is the model id, so model ids may contain the separator.
RawPart	A raw chunk of the provider stream. It appears only when RawChunks is set.
ReasoningDeltaPart	Reasoning delta.
RerankOptions	Rerank options.
SimulateStreamingMiddleware	Turns a non-streaming model into text deltas. Maps to simulateStreamingMiddleware.
SourcePart	A cited source.
StartPart	The full stream started. It is the first part of the stream.
StartStepPart	A step started. It comes before the parts of that step.
StepFinishPart	A step finished.
StepPerformance	Performance metrics for one step. Maps to StepResultPerformance.
StepResponse	Response information of one step. Maps to LanguageModelResponseMetadata & { messages, body? } on StepResult.response.
StepResult	One language-model step, including any tool results from that step.
StopCondition	Stops the tool loop. The default for generateText is IsStepCount(int) of 1.
StopWhen	Stop-condition factories. Maps to isStepCount, hasToolCall, and isLoopFinished.
StreamAbortContext	Passed to OnAbort.
StreamTextInclude	What streamText keeps of the request and the provider stream. Maps to StreamTextInclude.
StreamTextOptions	Options for CancellationToken).
StreamTextResult	Result of streamText. Enumerating Stream(CancellationToken) or awaiting Text consumes the same buffered events.
TextDeltaPart	Text delta.
TextStreamPart	A part of the streamText full stream.
Tool	A tool the model can call. Maps to tool().
ToolApprovalRequest	A tool call the caller may approve before it runs.
ToolApprovalRequestPart	A provider-executed tool call needs approval before the provider runs it.
ToolCallPart	A tool call.
ToolErrorPart	A tool call failed or was denied. A successful call is a ToolResultPart.
ToolExecutionEndEvent	Passed to OnToolExecutionEnd after a tool ran.
ToolExecutionOptions	The options a tool receives when it runs. Maps to the second argument of execute.
ToolExecutionStartEvent	Passed to OnToolExecutionStart before a tool runs.
ToolInputDeltaPart	A streamed tool-input fragment.
ToolInputEndPart	The model finished streaming tool input.
ToolInputStartPart	The model started streaming tool input.
ToolResultPart	A tool result.
TranscribeOptions	Transcription options.
TranslateOptions	Speech translation options.

Interfaces

Type	Summary
IAiClient	Core AI SDK client. Maps to the ai package functions.
IAiTelemetry	Telemetry hook invoked around generate and stream calls.
ILanguageModelMiddleware	Wraps doGenerate and doStream. Maps to language-model middleware.