AgentTalkResult

typesince 0.7.0not for built apps

Type: the state of an agent-talk run — returned identically by talkToAgent and getAgentResult.

import type { AgentTalkResult } from 'flowstack-sdk'

One shape covers both the inline answer and the polled job. status is the discriminator and the only field always meaningful: response/turns_used/cost_usd appear when done; message/poll_tool appear when running.

Signature

interface AgentTalkResult { status: 'done' | 'running' | 'error' | 'unknown'; job_id: string; agent_id: string; site_id: string; persona: string; response?: string; turns_used?: number; cost_usd?: number; routed?: boolean; message?: string; poll_tool?: string; error?: string }

Returns

Field	Type	Description
`status`	`'done' \| 'running' \| 'error' \| 'unknown'`	The discriminator. `unknown` means the job expired (1h TTL) or never existed.
`job_id`	`string`	The run id — pass to `getAgentResult` while `status` is `running`.
`agent_id`	`string`	The agent that ran.
`site_id`	`string`	The site the agent belongs to.
`persona`	`string`	The persona name that handled the task.
`response`	`string`	The answer. Present only when `status` is `'done'`.
`turns_used`	`number`	Agent turns consumed by the run.
`cost_usd`	`number`	Run cost in USD.
`routed`	`boolean`	True when the capability registry picked the target rather than the caller.
`message`	`string`	Human-readable status text. Present when `status` is `'running'`.
`poll_tool`	`string`	How to fetch the answer. Present when `status` is `'running'`.
`error`	`string`	Failure detail when `status` is `'error'`.

Warningresponse is optional. Reading it without first checking status === 'done' yields undefined on every run slower than waitSeconds — which is the normal case for research agents.

Warningstatus: 'error' and status: 'unknown' both arrive with a successful (ok: true) HTTP response. Transport success is not run success — branch on status, not on res.ok alone.

AgentTalkResult

Signature

Returns

See also