AIChatSendMessage¶

POST ai/chat/message

Queues a question and returns at once with a runId. It does not wait for the model — an answer takes tens of seconds and the platform closes a call after five, so the interface shows an indicator and then either listens to events or polls AIChatGetRun.

Access Control¶

Allowed: client, manager, admin. Only inside a conversation of the current session.

Request¶

RESTTCPSDK

POST request

POST https://{broker_domain}/ai/chat/message
Authorization: <JWT_TOKEN>
Content-Type: application/json

{
  "conversationId": "cnv_8f3a…",
  "content": "Why was position 500103 closed?",
  "requestId": "b1f0-4c2e-…",
  "route": "/backoffice/trades"
}

{
  "command": "AIChatSendMessage",
  "extID": "1",
  "data": {
    "conversationId": "cnv_8f3a…",
    "content": "Why was position 500103 closed?",
    "requestId": "b1f0-4c2e-…"
  }
}

const run = await platform.AIChatSendMessage({
  conversationId, content: 'Why was position 500103 closed?', requestId
});

Parameters¶

Parameter	Type	Description
`conversationId`	string	Conversation to ask in
`content`	string	The question, 1 to 8000 characters
`requestId`	string	Optional but recommended, up to 128 characters. See below
`route`	string	Optional. The current interface screen, for example `/backoffice/groups`
`accountLogin`	int	Account context. Required for a client session of an account owner
`symbol`	string	Optional. The instrument the user is looking at

requestId is generated by the interface. Repeating it returns the existing run with idempotent: true instead of paying for a second answer — protection against a double click, a network retry and a page reload during send.

route improves the answer. A knowledge document attached to that screen is ranked higher, so the same question asked from the groups screen and from the trades screen gets different, more specific answers.

Response Data¶

{
  "runId": "run_55b…",
  "messageId": "msg_1c7…",
  "state": "queued",
  "idempotent": false
}

messageId is the id of the user message that was just stored — use it to render the bubble immediately. The answer gets its own id, returned by AIChatGetRun in result.messageId.

Errors¶

Code	Error	Description
`402`	`AI_BUDGET_EXCEEDED`	The agent's daily request, token or cost limit is reached
`403`	`AI_ACCESS_DENIED`	The conversation belongs to another session
`404`	`AI_CONVERSATION_NOT_FOUND`	No such conversation in this scope
`409`	`AI_CONVERSATION_BUSY`	A run is already active in this conversation
`409`	`AI_AGENT_DISABLED`	The conversation's agent is disabled or not allowed in this chat
`503`	`AI_QUEUE_FULL`	The service queue is full
`503`	`AI_KNOWLEDGE_NOT_READY`	The knowledge index is still building

AI_CONVERSATION_BUSY is expected behaviour, not a failure: one run per conversation. Disable the input while an answer is being written instead of catching the error.