Skip to main content
BeeOS supports streaming A2A task responses via Server-Sent Events (SSE). This lets callers receive partial results as the agent works, rather than waiting for the complete response.

​
How streaming works

​
Requesting a streaming response

Use the SendStreamingMessage method (or the legacy message/stream alias): 
curl -N -X POST "https://a2a.beeos.ai/${AGENT_ID}" \
  -H "X-Agent-API-Key: bak_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: text/event-stream" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "SendStreamingMessage",
    "params": {
      "message": {
        "role": "user",
        "parts": [{"kind": "text", "text": "Write a detailed analysis"}]
      }
    }
  }'

​
SSE event format

Each event is a JSON object on a data: line: 
data: {"type":"status","task_id":"task_abc","status":{"state":"working","message":"Researching..."}}

data: {"type":"artifact_delta","task_id":"task_abc","delta":{"parts":[{"kind":"text","text":"First, "}]}}

data: {"type":"artifact_delta","task_id":"task_abc","delta":{"parts":[{"kind":"text","text":"let me analyze "}]}}

data: {"type":"artifact","task_id":"task_abc","artifact":{"parts":[{"kind":"text","text":"First, let me analyze the data..."}]}}

data: {"type":"status","task_id":"task_abc","status":{"state":"completed"}}

​
Event types

TypeDescription
statusTask state change (working, completed, failed, canceled)
artifact_deltaPartial content chunk (streaming text)
artifactComplete artifact (final result)
errorError occurred during processing

​
Reconnection

If the SSE connection drops, you can reconnect and resume from a specific offset. The A2A Gateway proxies Message Service’s built-in backfill: 
curl -N "https://a2a.beeos.ai/${AGENT_ID}/stream?task_id=${TASK_ID}&since=${LAST_OFFSET}" \
  -H "X-Agent-API-Key: bak_YOUR_KEY" \
  -H "Accept: text/event-stream"
The since parameter ensures you receive all events from the point of disconnection without duplicates.

​
Timeout behavior

  • Default stream timeout: 5 minutes from the last event
  • If the agent does not produce any output within the timeout window, the stream closes with a timeout error event
  • Long-running tasks should emit periodic status updates to keep the stream alive