Interface AgencyStreamResult<R>

Public stream result returned by agency().stream(...).

This exposes both low-latency raw streaming and finalized post-processing results so callers can choose the right trade-off for their UI or runtime.

Prefer:

  • textStream for raw token-by-token UX
  • fullStream for structured lifecycle events
  • text or finalTextStream for the finalized approved answer

textStream may differ from the finalized answer when output guardrails or beforeReturn HITL rewrite the result. finalTextStream and text always reflect the finalized post-processing output.

Example

const stream = team.stream('Summarize HTTP/3 rollout risks.');

for await (const chunk of stream.textStream) {
  process.stdout.write(chunk); // raw live output
}

for await (const approved of stream.finalTextStream) {
  console.log('Approved answer:', approved);
}

console.log(await stream.agentCalls);
console.log(await stream.text);
interface AgencyStreamResult<R> {
    textStream: AsyncIterable<string, any, any>;
    fullStream: AsyncIterable<AgencyStreamPart, any, any>;
    text: Promise<string>;
    usage: Promise<{
        promptTokens: number;
        completionTokens: number;
        totalTokens: number;
        costUSD?: number;
        cacheReadTokens?: number;
        cacheCreationTokens?: number;
    }>;
    agentCalls: Promise<AgentCallRecord[]>;
    parsed: Promise<unknown>;
    finalTextStream: AsyncIterable<string, any, any>;
    result?: Promise<R>;
}

Type Parameters

Properties

textStream fullStream text usage agentCalls parsed finalTextStream result?

Properties

textStream

textStream: AsyncIterable<string, any, any>

Raw live text chunks from the underlying strategy.

fullStream

fullStream: AsyncIterable<AgencyStreamPart, any, any>

Structured live + finalized event stream.

This includes raw text/tool/lifecycle events and also the finalized final-output event after post-processing completes.

text

text: Promise<string>

Finalized scalar text after guardrails, HITL, and parsing hooks.

usage

usage: Promise<{
    promptTokens: number;
    completionTokens: number;
    totalTokens: number;
    costUSD?: number;
    cacheReadTokens?: number;
    cacheCreationTokens?: number;
}>

Final aggregate usage for the streamed run.

Type declaration

  • promptTokens: number
  • completionTokens: number
  • totalTokens: number
  • Optional costUSD?: number
  • Optional cacheReadTokens?: number
  • Optional cacheCreationTokens?: number

agentCalls

agentCalls: Promise<AgentCallRecord[]>

Final per-agent execution ledger for the streamed run.

parsed

parsed: Promise<unknown>

Final structured payload, parsed from the finalized text (a beforeReturn approval's replacement when it carries one); resolves to undefined when structured output was not configured for the run or that text does not match the schema. A stream runs no validation retry.

finalTextStream

finalTextStream: AsyncIterable<string, any, any>

Finalized approved-only text stream.

Unlike textStream, this yields only the post-guardrail/post-HITL answer. For most runs it emits a single finalized chunk.

Optional result

result?: Promise<R>

The finalized result with its ledger, the same object generate() would return. Optional on the interface, so an object typed as it before this member existed still compiles; AgencyInstance.stream() returns it required.

Settings

Member Visibility

Theme

On This Page

textStreamfullStreamtextusageagentCallsparsedfinalTextStreamresult