TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Generative Mechanics · AGE 19—21 · FACTUAL · EVOLVING

Open-weights or API

Open weights run on your machines, and an API sends your prompt to a provider.

The idea

Some models give you their weights. You can download them, run them, fine-tune them, and — if the inference, the logging and every service around them also run on your machines — keep your data in-house. You can look at the weights; that does not let you read off what the model knows or why it behaves as it does. Others you reach only through a provider's API: your prompt travels to their servers and the answer comes back. Open weights give you control and the option of privacy. An API gives you ease and often the strongest models.

Why it matters

"Can we put the client's drawings into it?" is answered by the architecture, not the brand. Where the model runs decides where the drawings go.

See it in the studio

A hospital client's contract forbids their plans leaving the studio network. An API tool, however good, is out unless the provider's terms and hosting satisfy the contract. An open-weights model on a studio machine is in. The price is a machine, a person to run it, and a model a step behind the best.

Watch for this

"Open" used loosely. Released weights are not released training data, and some "open" licences restrict commercial use. Read the licence, not the adjective.

Try it

Take one real studio need with a privacy constraint. Write the two paths side by side: what runs where, who sees the data, what it costs per month, what you cannot do on each. Defend the pick in five lines.

Prove it

List what open-weights access lets you DO that an API does not — run locally, fine-tune, keep data in — and the reverse; tell open weights, open source, local and private apart; then match a studio's privacy and control needs to one path and defend it.

How it works

An API call sends your context to the provider's servers, and the provider's terms govern how long it is kept and whether it is used for training. Open weights run wherever you can supply the hardware, and can be fine-tuned locally (FINE-TUNE). The Open Source Initiative's definition of open-source AI asks for more than weights: the freedom to use, study, modify and share, with enough detail to rebuild the model. Most "open" releases do not fully meet it. EVOLVING: the capability gap between open and closed models moves every quarter, and so do the terms.

What this idea builds on

What this idea opens up

Sources

Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.

Age grows from 11 at the centre to 22 at the edge, and six sectors show the learning strands. Tab into the map and the arrow keys step from idea to idea, following the links where there is one. Enter opens the idea under the cursor, and E reads out its links and the reason recorded on each. Press slash for Search, question mark for the full key list, and Escape to leave. Open Ideas for the complete readable list, including what each idea builds on and what it opens up.

LOGIKA · RBDS AI LAB INDIA
ON-RAMP · AGE 11 · FIRST ENCOUNTERS, NOT GATES — IDEAS · — DEPENDENCIES
DONE
OPENS NEXT
SOLID — STANDS ON