Class: TemporalModel

googleAdkAgentsWorkflow.TemporalModel

A BaseLlm whose inference is durable under Temporal.

Swap a user's model: 'gemini-2.5-flash' for model: new TemporalModel('gemini-2.5-flash') — every model call inside the Workflow becomes a retryable, observable Activity, while the surrounding ADK agent loop replays deterministically.

Hierarchy

• BaseLlm

  ↳ TemporalModel

Constructors

constructor

• new TemporalModel(model, options?): TemporalModel

Parameters

Name	Type	Description
model	string	A model name registered in the ADK LLMRegistry (or resolvable by a custom modelProvider on the plugin).
options?	TemporalModelOptions	Per-model Activity configuration.

Returns

TemporalModel

Overrides

BaseLlm.constructor

Properties

[BASE_MODEL_SYMBOL]

• Readonly [BASE_MODEL_SYMBOL]: true

A unique symbol to identify BaseLlm classes.

Inherited from

BaseLlm.[BASE_MODEL_SYMBOL]

---

model

• Readonly model: string

Inherited from

BaseLlm.model

---

supportedModels

▪ Static Readonly supportedModels: (string | RegExp)[]

List of supported models in regex for LlmRegistry.

Inherited from

BaseLlm.supportedModels

Methods

connect

▸ connect(llmRequest): Promise<BaseLlmConnection>

Live bidirectional (BIDI) connections are not supported inside a Workflow — a long-lived two-way stream does not map onto the request/response Activity boundary. Outside a Workflow this delegates to the real model.

Parameters

Name	Type
llmRequest	LlmRequest

Returns

Promise<BaseLlmConnection>

Overrides

BaseLlm.connect

---

generateContentAsync

▸ generateContentAsync(llmRequest, stream?, abortSignal?): AsyncGenerator<LlmResponse, void, unknown>

Generates content for llmRequest. Inside a Workflow this proxies the model Activity; outside a Workflow it delegates to the real registered model so the same object is usable in non-Temporal contexts.

Parameters

Name	Type
llmRequest	LlmRequest
stream?	boolean
abortSignal?	AbortSignal

Returns

AsyncGenerator<LlmResponse, void, unknown>

Overrides

BaseLlm.generateContentAsync

---

maybeAppendUserContent

▸ maybeAppendUserContent(llmRequest): void

Appends a user content, so that model can continue to output.

Parameters

Name	Type	Description
llmRequest	LlmRequest	LlmRequest, the request to send to the LLM.

Returns

void

Inherited from

BaseLlm.maybeAppendUserContent