Connect your traffic
OpenTelemetry, Langfuse, Braintrust, LangSmith, Helicone, and Cloudflare AI Gateway.
The source picker is ordered by recommendation and labels unverified connectors as Preview. Preview means the parser has fixtures, but the pull has not been exercised against a live account end to end.
| Source | Maturity | How traffic arrives |
|---|---|---|
| OpenTelemetry | Supported; recommended | Your OTLP exporter pushes JSON or protobuf |
| API / webhook | Supported | Your application posts normalized batches |
| File upload | Supported | Import CSV, JSON, or JSONL from New Run |
| Langfuse | Preview | PeerLM polls the configured instance |
| Braintrust | Preview | PeerLM polls one explicit project |
| LangSmith | Preview | PeerLM polls one explicit project |
| Helicone | Preview | PeerLM polls request logs |
| Cloudflare AI Gateway | Preview | PeerLM fetches an exported log file |
Braintrust requires a project ID and LangSmith requires a project name; PeerLM refuses to guess. Polled sources paginate to a bounded end of a frozen import window.
Connection is not ingestion: saving credentials only creates the source. Wait until the Monitor reports received records before starting a traffic Run. If a preview connector does not import reliably, use a file export or OpenTelemetry rather than assuming its connection status proves end-to-end support.