Agent activity streaming

Each agent gets its own streaming token in the app. It is shown once when you create it and stored only as a hash, so keep it somewhere safe. Anything that can post JSON can stream; there is no session to manage.

Send events with an HTTP POST and an Authorization: Bearer header carrying the token. An invalid or revoked token returns 401.

Post a batch of events, up to 500 per request. Each event has a stable event_id (a redelivered id is ignored, so retries are safe), a run_external_id that groups events into one run, a type, an ISO 8601 occurred_at, and a payload.

curl -X POST https://app.superorgs.com/api/agent-stream/v0/events \
  -H "Authorization: Bearer sot_your_agent_token" \
  -H "Content-Type: application/json" \
  -d '{
  "events": [
    {
      "event_id": "evt_9f2a01",
      "run_external_id": "run_recon_0707",
      "type": "run.started",
      "occurred_at": "2026-07-07T15:04:00Z",
      "payload": { "title": "Nightly reconciliation" }
    },
    {
      "event_id": "evt_9f2a02",
      "run_external_id": "run_recon_0707",
      "type": "cost",
      "occurred_at": "2026-07-07T15:05:30Z",
      "payload": {
        "input_tokens": 18422,
        "output_tokens": 3110,
        "cost_micro_usd": 84200,
        "estimated": false
      }
    },
    {
      "event_id": "evt_9f2a03",
      "run_external_id": "run_recon_0707",
      "type": "run.finished",
      "occurred_at": "2026-07-07T15:06:12Z",
      "payload": { "summary": "Reconciled 1,204 rows, no exceptions." }
    }
  ]
}'

A cost event carries input_tokens, output_tokens, cost_micro_usd, model, and estimated. SuperOrgs folds these into the run's real metered cost.

• Live feeds, org-wide and per agent, newest run first.
• Real metered cost per run in micro-USD, not an estimate.
• The full event timeline for any run: messages, tool calls, and cost.
• A one-line Orion summary written for each run.
• Follow an agent in Slack and get its runs where you work.
• Error forensics: open a failed run and see exactly what happened.

The whole stream is queryable from your assistant. Ask what an agent did, what it cost, or why a run failed, right from Claude, ChatGPT, or Cursor. See Connect your AI assistant.