Real-time and streaming

OmniDimension is a real-time product at the voice layer and an event-driven product at the API layer. This page maps each real-time need to the right mechanism.

Streaming voice on live calls

Every call runs on a streaming pipeline: caller audio is streamed to speech-to-text, the LLM response is generated and streamed to text-to-speech, and synthesized audio streams back over the call, with sub-200ms end-to-end latency. The same streaming pipeline powers phone calls, the web voice widget, and WhatsApp voice.

You do not manage this stream yourself. Dispatch a call with the API, SDK, or MCP server and the platform handles the audio streaming.

The same applies to the embeddable web voice widget: it streams audio between the visitor's browser and the agent in real time, and agents can run live web search inside a conversation when enabled, so answers can include current information mid-call.

Live call monitoring

The dashboard streams active calls in real time: watch conversations as they happen across your agents. This is a dashboard surface (backed by a WebSocket service), not a public API; for programmatic visibility into running activity, use the bulk-call live status endpoint below or fetch call logs after completion.

Webhooks: push delivery for call events

For machine-to-machine delivery of call results, use webhooks. When a call completes, OmniDimension POSTs your configured payload to your endpoint: transcript, summary, sentiment, extracted variables, and the recording URL.

• Setup guide: Post-call delivery
• Configure under Post call on your agent: choose Webhook, paste your endpoint URL, and define a custom payload with call variables.
• Test with Webhook.site or RequestBin to inspect real payloads before pointing at production.

Webhooks are the recommended pattern for agents and integrations that need results as soon as a call ends, instead of polling call logs.

Webhook is one of several post-call delivery channels: the same completion event can also be delivered to Email, Salesforce, Slack, HubSpot, or WhatsApp Cloud, configured per agent.

Polling: live status for bulk campaigns

Running campaigns expose a lightweight live status endpoint designed for repeated polling. It returns aggregated counts in one cheap query:

curl https://backend.omnidim.io/api/v1/bulk-call/{bulk_call_id}/live-status \
  -H "Authorization: Bearer $OMNIDIM_API_KEY"

{
  "status": "success",
  "bulk_call_id": 42,
  "campaign_status": "in_progress",
  "summary": {
    "total_contacts": 500,
    "queued": 120,
    "in_progress": 8,
    "completed": 350,
    "failed": 12,
    "busy": 6,
    "no_answer": 4
  }
}

Poll every few seconds while a campaign runs. For individual call results, prefer webhooks over polling call logs.

Server-sent events (SSE)

The REST API does not currently expose SSE or chunked streaming responses; API responses are standard JSON. Real-time delivery is covered by the mechanisms above: streaming voice inside calls, webhook push for call events, and the live-status polling endpoint for campaigns. If your use case needs SSE specifically, contact us.

Service status

Live availability for the API, omnidim.io, SIP infrastructure, and the MCP server: status.omnidim.io. Agents can read it as JSON: https://status.omnidim.io/api/status-page/heartbeat/all (heartbeats and uptime per service).

Choosing the right mechanism