AI models whose trained parameters are published so anyone can download and run them. For agencies, they offer a way to keep client material on infrastructure you control.
Also known as open weights, downloadable models
An open-weight model is one whose trained weights, the numbers that make up the model, are released for download. You can run it on your own hardware or a cloud server you control, and often fine-tune it. This is different from a closed model reached only through a vendor’s API.
“Open weight” is not the same as fully open source. The training data and code may not be shared, and the license can limit commercial use, so the terms matter as much as the capability.
Agencies sit on confidential client data, and where a model runs decides who sees it.
Confidential work stays in-house. Unreleased product details and client financials never leave your environment when the model runs on your own infrastructure.
Costs are predictable at volume. High-volume, repetitive tasks such as tagging assets or summarizing reports can be cheaper on a model you host than on per-call pricing.
There is a capability and upkeep tradeoff. Open models can lag the best closed ones on hard tasks, and someone has to maintain the setup, secure it and check the license.
A pharmaceutical client’s contract bars sending campaign drafts to third-party AI services. The agency’s tech lead sets up an open-weight model on a private server to summarize research and tag creative variants, with access limited to the account team. Final copy polish still goes through a stronger closed model, using only non-confidential text. The client signs off because the sensitive material never leaves the private server.
The automations and agents module of the workshop teaches you how to build AI workflows that compress the busywork without taking the craft out of the studio.