Platform/Ingestion
Production Call Ingestion
Multiple paths to observe production calls without hosting your agents: webhooks, connectors, SDKs/OTel, batch and streaming.
Design principles
- High-volume async processing with backpressure and partitioning
- Deduplication via idempotency keys; failure recovery via DLQ + retries
- Normalize to canonical model; retain original raw payload for investigation
- Every call gets a global internal ID mapped to all provider IDs
Webhook ingestion
20 event types · HMAC · idempotency · rate limits · DLQ · audit
OpenNative connectors
Voice AI + telephony providers via adapter framework
OpenSDKs & OpenTelemetry
9 languages · traces · metrics · Voice AI semantic conventions
OpenBatch import
JSON · JSONL · CSV · Parquet · S3 · Kafka · Kinesis · Pub/Sub
OpenStreaming ingestion
WSS · WebRTC · SIPREC · RTP · Kafka · backpressure · partitions
OpenUnified event model
Vendor-neutral entities · global call ID · raw payload retention
Open