Features
The operational layer your agents were missing.
Generation is a solved problem. Nebula owns the hard half: keeping agents durable, observable, and safe in production.
Durable runs
Runs are checkpointed after every step and stored durably. Redeploy mid-run, survive a crash, or retry a flaky tool, state is never lost, and nothing runs twice by accident.
Live tracing
A timeline of every tool call, prompt, token, and latency number, streamed live. Replay any historical run against a new prompt or model and diff the outcome side by side.
Guardrails
Declarative policies sit between generation and action. A call that crosses a threshold pauses for a human on web or mobile, then resumes cleanly with their decision injected.
Evals in CI
Point Nebula at an eval suite and it scores every change in CI. The build goes red when quality regresses, with a per-case delta so you can see exactly what got worse.
Scoped tools
Expose only the tools each agent needs, scoped per client with hard permission boundaries. Every call is logged to an immutable audit trail you can export.