Spanrail: self-hosted LLM observability with per-call timing

See where every LLM call spends its time.

Traces, evaluations and prompts in one self-hosted place, with DNS, connect, TLS and first-byte timing for every call that passes through the gateway.

One parcel: one event, one rule for each kindThe press: the workerA resend, turned awayYour app or the gateway sends a batch of events: one to start the trace, one to start the generation, one to finish it. Click to replay. The events are illustrative; the steps are what Spanrail's ingestion pipeline does.

Runs on your own infrastructure. Your data stays with you.

Spanrail

See where every LLM call spends its time.

Sign in