Embabel Agent Documentation/com.embabel.agent.api.common.nested/TemplateOperations

TemplateOperations

interface ~~TemplateOperations~~ : PromptRunner.Rendering

Deprecated

Use PromptRunner.Rendering instead

Replace with

import com.embabel.agent.api.common.PromptRunner.Rendering

PromptRunner.Rendering

Llm operations based on a compiled template. Similar to com.embabel.agent.api.common.PromptRunnerOperations, but taking a model instead of a template string.

Functions

createObject

abstract fun <T> createObject(outputClass: Class<T>, model: Map<String, Any>): T

Create an object of the given type using the given model to render the template and LLM options from context.

createObject

inline fun <T> PromptRunner.Rendering.createObject(model: Map<String, Any>): T

generateText

abstract fun generateText(model: Map<String, Any>): String

Generate text using the given model to render the template and LLM options from context.

respond

open fun respond(conversation: Conversation, model: Map<String, Any>, onFailure: (Throwable) -> AssistantMessage): AssistantMessage

Safely respond to the user in the conversation using the rendered template as system prompt, returning an error message if something goes wrong. Cannot throw an exception.

respondWithSystemPrompt

abstract fun respondWithSystemPrompt(conversation: Conversation, model: Map<String, Any> = emptyMap()): AssistantMessage

Respond in the conversation using the rendered template as system prompt.

respondWithTrigger

abstract fun respondWithTrigger(conversation: Conversation, triggerPrompt: String, model: Map<String, Any> = emptyMap()): AssistantMessage

Respond to a system-initiated trigger using the rendered template as system prompt. The trigger prompt is appended as a user message to the LLM call but not stored in the conversation.

withTemplateRenderer

abstract fun withTemplateRenderer(templateRenderer: <Error class: unknown class>): PromptRunner.Rendering

Set the template renderer to use for rendering templates in this mode. By default, this Renderer will use the platform's default TemplateRenderer, but this allows for custom renderers to be used on a per-rendering basis.