Skip to main content

Send a message to the assistant and stream its run

POST 

/plugin-api/ai/ops/chat

Server-sent events, like /chat/stream, plus a tool_progress frame for each host as its command finishes, a targets frame when the assistant changes the targets and a client_action frame when it asks the browser to open a view. reasoning frames carry the model's thinking apart from its tokens, a usage frame follows every model call with its tokens, the context window, its source (override, provider, guess or default) and the tokens kept for the reply, and a compacted frame marks where the history was summarised to fit. The run goes on in the background: closing the stream only stops watching it, and /ops/conversations/{id}/stream picks it up again. There is no step limit; the run ends when the model answers, it is stopped, or it repeats one call with one result five times. One run at a time per conversation. hostIds become the conversation's targets; leave them out to keep the ones it has. Targets, and the tools that run commands on them, need ai.ops.use; without it a non-empty hostIds is refused and the CLI cannot exec on hosts. page is what the user is looking at.

Request​

Responses​

An event stream.