AI models
Choose which model creates dashboards, edits them, answers data questions and answers viewers.
Flitch uses AI in four places, and each one can run on a different model. Open from Settings → AI models.
The setting belongs to the space, not to you: credits come from the shared pool, so a stronger model spends everyone's allowance faster. Any member can see what is in use. Only an admin can change it.
Saves as you choose. A change applies to work started after it, not to anything already running.
The four areas
Creating dashboards
Writes the pages, the navigation and the theme when a dashboard is first built. This is the largest single piece of AI work Flitch does, so it is where a change in model shows up most in both quality and cost.
Default: Claude Sonnet 5.
A few smaller steps keep their own models whatever you choose: planning what the dashboard should contain, naming it, generating sample data, and the repair pass that fixes broken code. Each is picked for its job, and together they are a small share of what a dashboard costs.
Editing dashboards
Handles the changes you ask for in the editor chat.
Default: chosen per request. A small, cheap model reads what you asked for and picks: a label change goes somewhere cheap, a restructure goes somewhere capable, and a bug report is never sent to the cheapest tier. Naming a model here turns that off and uses your choice every time, which costs more on trivial edits and less on hard ones.
Data assistant
Answers questions about your data on the Data page, and proposes changes to it.
Default: Claude Sonnet 5.
Ask AI on published dashboards
Answers questions from people viewing a dashboard you published. Those answers are billed to your credits, so this is the one to think about if a dashboard is shared widely. Turning Ask AI on or off per dashboard is in the publish settings.
Default: Claude Sonnet 5.
Where else you can change it
The same choice appears next to the work itself, and each of those links back here:
- Creating: the Settings panel on the prompt input.
- Editing: the + menu in the editor chat, under Model.
- Data assistant: the composer bar on the Data page.
- Ask AI: the icon beside the Ask AI toggle, in the publish menu and the Publishing tab.
What a change costs you
Credits come off what a request actually costs to run, not a flat price per request. So the model you pick decides how far your allowance goes: a stronger one does better work and spends credits faster, and a cheaper one stretches the same allowance further.
Nothing about your plan changes. Only the rate you spend it at.
Haiku is offered for editing, the data assistant and Ask AI, but not for creating a dashboard: writing a full set of pages is beyond what it does well.
Watch it for a few dashboards
The difference between models is easiest to judge on your own work rather than in the abstract. Change one area, build a couple of dashboards, and compare both the result and what your balance did.
Who processes your data
Choosing a model decides which company receives your prompts, dashboard structure and any data samples included in them. Anthropic, OpenAI, Google and xAI are all listed as sub-processors, with the terms under which each holds that data, at flitch.io/sub-processors.
If your organisation restricts which providers may process its data, set the model here rather than relying on the default.