Skip to main content

Choosing AI providers and models

With Floxar, how your work is done does not belong to any AI provider. Your processes live in Floxar as flows, and the record of how they ran lives in Floxar as trails. The AI that runs a step is your choice, and you can change it whenever you like: a provider and model today, a different one tomorrow, or a different model for different kinds of work.

Your processes outlive any model​

When a procedure is written into the instructions of one AI agent, it is tied to that agent and, in practice, to the model it was written and tested for. Changing models means rewriting and retesting the instructions.

In Floxar, the procedure is the flow. An agent's instructions stay short and general (find the flow, follow it, record what happened), so switching the model or the provider behind an agent leaves your flows, your required answers, your hand-offs and your history exactly as they were. Your flow library and your trail history carry across every change of model. See Where Floxar fits in an AI stack.

Choosing a model for each kind of work​

Not every process, or every step, needs the most capable model. Floxar gives the agent that runs your work what it needs to choose:

  • By step: complexity. Every bit has a complexity setting, Low, Medium or High, saying how much reasoning effort an AI agent should invest in that step; Medium by default. An agent can use a fast, inexpensive model for Low steps and a stronger model for High ones. Set it in the bit's properties; see The bit editor.
  • By process: priority and categories. Every flow has a priority and can carry categories of your own choosing. An agent can choose its model from them: for example, the most capable model for flows you mark critical, a private model for flows tagged with sensitive data, and a lower-cost model for routine work.
  • By your own rules. Because the choice is made by your agent, you decide the rules: which providers you trust, where data may go, and how much each kind of work may cost.

The Operator, once available, reads a step's complexity as its guide to which model to use. See The Operator.

Smaller models do more​

Each step in a flow gives the agent only what that step needs: its instructions, the answers it asks for and the paths that lead on. A narrow, well-defined task is something a smaller or locally hosted model can often handle well, where the same model would struggle with a whole procedure and a pile of documents. The better your flows are written, the more of your work can run on smaller models; see Writing effective bits.

Mixing providers, and people​

Different agents can run different flows, each on the provider that suits it, and a single run can pass between agents and people as the process requires. The trail records every step the same way whoever performed it, so you can compare how well different models handle the same work before deciding where to use each.

Floxar's own AI features​

The AI Assistant and the Flow Designer are run by Floxar, which chooses the models behind them. The agents that run your work, and the AI clients your people connect, use the providers and models you choose.

Last reviewed: 2026-10-06