AI presets
Bricks Technical Documentation
Which AI model, run in which country, under which legal chain, processes your studio AI — a small, curated, honestly-tagged choice, and why Bricks deliberately offers more than one.
A preset is the answer to a single question: when Bricks's studio AI does its work for you, which provider runs the model, on whose hardware, in which jurisdiction? You pick a preset in the AI Behaviour panel; Bricks holds the contract, the credentials and the data-processing agreement behind it. You choose; Bricks contracts.
A preset is not a "bring your own key" field and not a free-text endpoint. It is one reviewed (provider, endpoint, hosting region, sub-processor chain, jurisdiction) tuple that Bricks resells under a single relationship. The catalog is deliberately small because each row is deliberately audited — the goal is a choice you can trust without having to audit it yourself.
The default is honest about where it runs
Every tenant starts on the platform default. It is the highest-quality option Bricks can offer for the studio AI's multi-step work today, and it is tagged with its real jurisdiction wherever you see it — the catalog row, the description that opens when you click the preset name, and the record stamped onto every AI message. There is no sovereignty language dressing up a choice that cannot substantiate it: if a preset runs under US law, it says so; if it runs under EU law, it says so.
Where the default's jurisdiction does not fit your needs, you switch. That is the entire point of having more than one row.
Why Bricks maintains more than one preset
Offering a choice of preset is a feature, not a hedge. The reasons a customer needs the choice fall into four families — and most real decisions mix several of them.
Technical reasons
- Capability, latency and cost differ by workload. No single model is best at everything; a preset that is excellent for one kind of work may be over- or under-powered for another. A plural catalog lets the right model do the right job.
- Continuity and availability. A single provider is a single point of failure. If one provider's service is degraded, or becomes unavailable for reasons outside anyone's control, a second load-bearing preset is how your work keeps moving instead of stopping. Availability is part of security, not separate from it.
- Portability. Open-weight models can run on more than one host, so a preset built on them is not locked to a single company's uptime or pricing. That portability is what makes "keep working no matter what happens upstream" a real promise rather than a hope.
Cultural reasons
- Less monoculture. A catalog anchored on a single model lineage inherits that lineage's cultural defaults — its language centre of gravity and its sense of what a "normal" user looks like. Offering models of different origins lets your frame — your language, your context — set the terms, instead of the platform's.
- Language and register. Different models have different strengths across languages and writing styles. The choice is yours to make against the audience you actually serve.
Political reasons
- Models carry their editors' positions; Bricks does not. Bricks is non-partisan. It takes no position on any state's political matters, and it does not moralise about which preset you pick. Different presets exist precisely so that the position — if any — is yours to hold in your work, with full transparency over how anything was generated.
- There is one line Bricks does not stay neutral about. Bricks enforces a floor of inalienable rights — the right to live, to believe, to take part in the economy, to hold privacy. That floor protects people who use what you build: it is enforced between an abused end user and the organisation that used Bricks, through the platform's audit trails, never as a stance on any government's politics. Bricks is not a tool to promote hate or to abuse others' rights; it is a tool on which you can honestly state a position and even gather opposing views — with the model and jurisdiction behind every contribution made visible.
Legal reasons
- Jurisdiction and data-residency exposure. Where inference physically runs decides which legal regime can reach the data — regardless of where a company's letterhead reads. If your procurement rules, your regulator, or your own risk assessment require an EU jurisdiction (or exclude a particular one), the preset you pick is how you satisfy that.
- One contract, one DPA. Because presets are resold rather than self-connected, the data-processing agreement, the sub-processor disclosure and the support relationship are Bricks's to hold — you do not negotiate a separate contract per model.
How you choose, and how it stays transparent
- Honest tags. Each preset shows its jurisdiction and a plain-language description of its hosting chain and trade-offs before you select it.
- A consent step on every switch. Changing preset opens a short modal that tells you what the next AI turn will send to the new provider (recent conversation context and the ability to read your apps, workflows and tables through the studio AI's tools) and what it will not (the studio AI cannot read the rows of your tables, whichever preset you are on — only what you yourself paste or attach in the chat reaches the provider. Before it does, every message and its attachments are checked for personal data by Mistral Small, hosted by Scaleway in France, whatever your preset; a message that carries some is held back and reaches your provider only if you send it anyway). Each switch is a fresh authorisation of a new processor, not a one-time setting.
- Per-message provenance. Every message records the preset it was sent under, and every AI answer also records the model actually used — stamped at the time it ran, so your history is not re-labelled when you switch later; the jurisdiction shown is the one that preset is linked to. If a preset ever falls back to another model for continuity, the recorded model shows it.
Your reasons are your own
Sovereignty, taken fully, includes the right to choose or exclude a provider for reasons you do not have to explain — a concern that may be well-founded or simply yours. Bricks is built to honour that: you are not required to justify an exclusion, and Bricks does not record a why behind your choice that could later be used to profile you. Pick the AI you are willing to trust, for reasons that are yours.
What a preset is not
- Not a perfect sovereignty guarantee. Every hosting chain eventually reaches hardware vendors, networks and power grids with their own jurisdictional attributes. Each preset's notes describe what is verifiable about its chain and what is not; the catalog is honest about the limit rather than papering over it.
- Not bring-your-own-key. The catalog is resold-only by design. Sovereignty that only works for customers able to provision and govern their own AI providers is not sovereignty for the audience Bricks is built for.
- Not a moral ranking. The default is a defensible choice, not a failed sovereignty test; the EU option is there for customers whose context requires it, not as a badge of virtue.
- Not a promise that every preset is equally capable. Presets differ in quality on different work, and a continuity fallback means the platform keeps working, not you get an identical answer. The trade-offs are disclosed, never hidden.
The full reasoning behind how Bricks curates this catalog — why it is small, why it is plural, and why the default is tagged rather than dressed up — is set out in the Curated Pluralism whitepaper.