Ready-made, or made to measure.

For work that general models already do well: drafting, summarizing, answering from your own documents, filling in forms. The craft is in the fitting.

• We map the workflow and the data it touches.
• We choose the model and connect it to your documents and tools.
• We agree a test and measure the model against it.
• We deploy on vLLM in your cloud or server room, or on your Macs with Melix.
• We keep it running and bring in better models as they appear.
• If your data cannot leave your building, we do the work offline on site.

For work that depends on what only your company knows (its terms, procedures, formats and software), or when you need a smaller model that runs on your own hardware.

Everything in Ready-made, plus:

• a dataset built from your material, with a dataset card;
• fine-tuning and preference tuning on your people's real tasks;
• continued pre-training, when your field's vocabulary is missing from general models.