Trace timeline
Nested spans for every model call and tool, with latency and cost on each.
Nocturne traces every tool call, prompt and decision your AI agents make, so debugging takes minutes instead of days.
Drop in one SDK call. Nocturne captures the full tree of reasoning, tools and tokens, and turns it into something a person can read.
Nested spans for every model call and tool, with latency and cost on each.
Re-run any trace against a new prompt or model and diff the outcome.
Score runs automatically and catch regressions before they ship.
Budgets per agent, per user, per day, with alerts before you blow through them.
Self-host in your VPC. PII is redacted at the edge.
Five minutes to your first trace. No credit card.