Step one

Choose a model for your machine

Model choice matters more than it looks, and memory is the first filter. Pick the card that matches your hardware.

A server with 512 GB of memory glm-5.2

Index 88.3 Drafted 10 of 10 about 309 GB at 4-bit

Within about 3.3 points of the best closed-weight model measured, with nothing leaving the building. This is the strongest local option and it needs server hardware.

Show the commands

A workstation with 64 GB gemma4:31b

Index 79.8 Drafted 9 of 10 about 30 GB, flat

The best workstation-tier model measured. It failed the largest specification in the corpus (645 KB); for one that size, use gemma4:26b below, which drafted all ten.

Show the commands

A desktop or laptop with 32 GB gemma4:26b

Index 65.0 Drafted 10 of 10 about 20 GB, flat

Drafted every specification in the corpus, including the 645 KB one, and is honestly behind on drafting quality. Memory stays flat as the specification grows, which is what lets it finish the large ones.

Show the commands

A laptop with 16 GB gemma4:12b

Index 62.9 Drafted 9 of 10 about 12 GB, flat

Marginal, and uneven across specifications. Workable for trying the tool; treat its output as a rougher first pass than the tiers above.

Show the commands

Setting up gemma4:26b

Four commands and you are drafting

Install Ollama

Get it from ollama.com/download and start it. The desktop app starts the server for you; otherwise run ollama serve.

Pull the model

A large download, once.

ollama pull gemma4:26b

Get the binary

From the download page, then make it executable.

Draft

Point it at your specification and the filed parent claims.

DRAFT_MODEL=gemma4:26b ./continuation-drafter draft \
  --spec spec.txt --parent parent-claims.txt

Not enough memory, or would rather not install a model?

Run it through your own provider account instead. That sends the specification to that provider, so it suits published material, and for anything unpublished it is a judgment you make per matter. How that works.