The best model for a task changes every few months, and so does its price. An AI agent platform tied to one provider ties your processes to that provider's roadmap. In ATX Agent Hub the model is a setting on a workflow step — change it, test it and promote it like any other change.
Model access runs through Azure AI Foundry and Google Vertex Model Garden, the two enterprise AI gateways most organisations already have agreements with.
A single workflow can classify with a small, fast model, reason with a frontier model and summarise with a cheaper one. Because each step records its model and token cost on every run, the trade-off between quality and cost is visible rather than guessed.
The Model Comparison node sends the same request to several providers and selects the best answer. It is useful both in production, for high-stakes answers, and during evaluation, to decide which model a workflow should use.
Changing a model is a new workflow version. Test suites run the golden cases against it, the regression dashboard compares versions, and promotion to production needs an independent approver. A model upgrade becomes a controlled release rather than a silent behaviour change.
Token use and cost are tracked per run, workflow, project, user and model, with forecast and burn-rate dashboards. Soft caps warn and hard caps block new runs, so a model that turns out more expensive than expected cannot run up a bill unnoticed.
Models available through Azure AI Foundry and Google Vertex Model Garden — including OpenAI GPT, Claude, Gemini, Llama, Mistral, DeepSeek and Cohere — and self-hosted models in on-premise deployments.
Yes. The model is chosen per workflow step, and every run records which model each step used and what it cost.
No. Models are reached through your Azure AI Foundry or Google Vertex AI account, under the cloud agreements you already have.
This site needs JavaScript for the full experience. Every page above is reachable without it.