Managed intelligence

The add-on operates the same authoring loop the open engine runs — planning, drafting, verification and repair — with operated capacity. It can use your own provider keys (BYOK) or the managed catalog.

What is included

Budgeted generation per job with reservation before dispatch and settlement after.
Deterministic checks first: model output is verified against extracted facts before publication.
Repair attempts bounded per job; a job that cannot be verified stops and says so.
Parallel editor plans for AI, visual and configuration edits against the same document program.

How it is evaluated

We evaluate the managed strategy against the full open-source workflow using the same base model, the same sources, the same tool rights and a comparable budget, scoring task success with an independent harness rather than the model grading itself. Intervals that overlap never establish superiority, and published results include model versions, spend and latency. Until that benchmark runs, no quality or speed advantage is claimed.

See pricing Read benchmarks policy

Actions

Try on a sample
Opens the real renderer playground on synthetic blocks: no account, no provider and no cost. It demonstrates rendering, not managed intelligence output.
Connect BYOK
BYOK is available on every plan including free, and the self-host guide shows the local environment variables; managed model spend is optional and metered separately.