← All documentation

AI drafts

The pipeline

A draft is written in steps: an outline with the figures, the claims, an introduction to the description, one section per figure, variants, and the abstract. Each step is saved, so a draft that was interrupted continues where it stopped. Office (DPMA, EPO, USPTO, PCT) and language (German, English) decide format and wording.

Which model

The local model on the firm's server is the default where one is installed; without a local model, no external model is used until one is chosen for the application, the client or as the firm default. Administrators can add providers - their own OpenAI-compatible servers or cloud services - and decide per client which models are allowed; a file can pick one of them. Every call is logged with model, user, file and duration.

Firm guidelines

Settings → AI guidelines holds the firm's rules as Markdown: checklists, structure, example claims, glossaries - per task, office and language, with versions. They are added to every request, and a preview shows exactly what the model receives.

Queue

Drafts and claim charts run one after the other; revisions and other short tasks have their own queue and are answered between the steps of a running draft. The file shows how many jobs are ahead and how long drafts usually take.