Stream Agent Response (SSE)

Ask an agent a question and receive the answer in real time as a streamed agent response. You receive progress updates and reasoning steps as they happen. Use this when you need to display live agent activity or react to events as they arrive.

Pass ?agent_version=staging to chat with the staging version of the agent.

Usage Limits

Subject to your organization's spending limit. Requests fail with 429 Quota limit exceeded if you hit the limit.

Current Version

2.3.0

Changelog

2.3.0: Added commands.required_skill_ids to apply specific connected skills to a request.

2.2.0: Added an SSE id to every event you can resume from /resume-events. Existing clients can ignore it. Added keep_generating: set it to true and the response keeps generating when your connection drops, so you can resume it later. Use /stop-requests to end it early.

2.1.1: Added param_override inside overrides. Pass {"effort": "high"} to select a reasoning-effort label.

2.1.0: Deprecated created_by and caching_enabled from the request body. The requesting user is logged automatically from the API key; configure caching with the Update an Agent endpoint.

Responses

🟢200OK

text/event-stream

Server-Sent Events, in order. Every event's data has a step.

event	`step`	What it means
`start`	`start`	Always the first event. Carries `run_id` (needed to stop or resume the response) and `trace_id`.
`status`	varies, e.g. `reasoning`, `assistant_start`, tool steps	A progress update. May carry `output` (e.g. a title and description, or a todo list).
`token`	`subagent_stream`	Progress narration from a sub-agent while it works. Not part of the answer: show it separately (e.g. as a live status line), don't append it to the answer.
`token`	any other `step` (e.g. `model_stream`)	A piece of the answer. Append `delta` to the answer text.
`ping`	`ping`	Keepalive after a quiet gap. Carries no data; ignore it.
`error`	`error`	Something failed. `output.error.code` is the machine-readable code and `output.error.details` the message to show. Always followed by `end`.
`end`	`end`	Always the last event. `output.status` says how the response ended: `completed`, `stopped`, `errored`, or `disconnected` (only seen when resuming). On `completed`, `output` holds the full answer.

Event id: each event you can resume from has an id: line holding its event id (e.g. id: 1789768688301-4), which marks its place in the response. start, ping, and every event of a response that can't be resumed have no event id. The examples show a response that can be resumed.

Bodytext/event-stream

🟠401Bad Request

🟠403Parameter Error

🟠422Server Error

🟠429

🔴500

curl --location --globoff 'https:///response-service/api/v2/agents//responses/stream?agent_version=' \ --header 'Ocp-Apim-Subscription-Key: ' \ --header 'Authorization: Bearer ' \ --header 'Content-Type: application/json' \ --data '{ "attachment_metadata": { "corpus_id": 1, "doc_ids": [ 1 ] }, "chat_thread_id": 1, "history": [ { "bot": "Paris is the capital of France.", "user": "What is the capital of France?" } ], "is_file_attached": false, "keep_generating": false, "overrides": { "param_override": { "effort": "high" }, "thinking_mode": "instant" }, "query_source": "web", "share_token": "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx", "user_query": "and what about the weather?" }'

event: start data: {"message": "Thinking...", "step": "start", "run_id": "5f0c6a1e-8a2b-4c3d-9e4f-1a2b3c4d5e6f", "trace_id": "2c1f6e0a9b8d7c6e5f4a3b2c1d0e9f8a"} event: status id: 1789768688301-1 data: {"message": "", "step": "reasoning", "output": {"title": "Searching documentation", "description": "Retrieving relevant entries from the corpus (knowledge base).", "source": "supervisor"}} event: status id: 1789768688301-2 data: {"message": "Todos written.", "step": "tools_write_todos_on_tool_end_end", "output": {"todos": [{"id": "1", "content": "Summarise the travel policy", "status": "pending"}]}} event: token id: 1789768688301-3 data: {"delta": "Checking the travel policy for per-diem limits.", "step": "subagent_stream"} event: ping data: {"step": "ping"} event: token id: 1789768688301-4 data: {"delta": "Here is a summary", "step": "model_stream"} event: end id: 1789768688301-5 data: {"message": "Finished", "step": "end", "output": {"answer": "Here is a summary of the travel policy...", "thread_id": 1, "agent_response_id": 1, "corpus_id": null, "fork": null, "references": [], "artifacts": [], "chat_thread_name": "Travel policy summary", "blocked": false, "cache_hit": false, "data_points": [], "citations": [], "trace_id": "2c1f6e0a9b8d7c6e5f4a3b2c1d0e9f8a", "attached_file_metadata": [], "status": "completed"}, "success": true}