Interface LoopContext

Execution context provided to the LoopController by the caller. Abstracts away the underlying LLM/GMI implementation so the loop logic remains provider-agnostic.

interface LoopContext {
    generateStream: (() => AsyncGenerator<LoopChunk, LoopOutput, undefined>);
    executeTool: ((toolCall) => Promise<LoopToolCallResult>);
    addToolResults: ((results) => void);
}

Properties

generateStream executeTool addToolResults

Properties

generateStream

generateStream: (() => AsyncGenerator<LoopChunk, LoopOutput, undefined>)

Async generator that streams chunks during a single LLM inference pass. Must return a LoopOutput as its generator return value (the value passed to the final done: true result from .next()).

Type declaration

    • (): AsyncGenerator<LoopChunk, LoopOutput, undefined>

    • Returns AsyncGenerator<LoopChunk, LoopOutput, undefined>

executeTool

executeTool: ((toolCall) => Promise<LoopToolCallResult>)

Execute a single tool call and return its result. Implementations should never throw — instead return a result with success: false and a populated error field.

Type declaration

addToolResults

addToolResults: ((results) => void)

Feed tool results back into the conversation so the next generateStream call has access to them. Typically appends tool messages to the message list.

Type declaration

Settings

Member Visibility

Theme

On This Page

generateStreamexecuteTooladdToolResults