Model providers

Any model. Your choice. Swappable anytime.

22 providers, one runtime. Agents and workflows are provider-agnostic — change the model behind an operation without touching the operation.

Choosing between them

The question worth asking is not which provider is best but which step deserves which model. A single workflow routinely mixes them: a frontier model for the judgment call that decides the outcome, a fast-inference tier for high-volume classification, and a local model for anything touching data you would rather not send anywhere. Because AGNT normalises tool calling across every provider on this page, that split is a setting on each step rather than three separate builds.

The practical consequence is that provider choice stops being a commitment. You can start on whatever you already have a key for, measure what your workflows actually cost and how often they are right, and move individual steps as the evidence arrives — without rebuilding the operation around a new vendor. Teams with strict data requirements run the same workflows entirely on local models through Ollama or LM Studio, at which point no inference leaves the machine at all.