AI · Models
Use any model. Change it without changing anything else.
Register any OpenAI-compatible endpoint (hosted, open-source or your own server) with your own key, and give each model a profile name. Prompts, pipelines and agents refer to the profile, so switching models or adding a fallback is one change in one place.
What you can do
Bring your own model
OpenAI-compatible endpoints, including OpenAI, Mistral, Hugging Face endpoints, Ollama, vLLM and your own server.
Chat and embeddings
The same providers serve chat and embedding models, so retrieval is not tied to one vendor.
Profiles
A named profile such as default points at a model with its settings. Everything else uses the name.
Fallback
If a profile fails, its fallback profile answers instead.
Test in one click
Test a model and see the latency or the exact error.
Cost and usage
Enter list prices and see calls, tokens, latency and cost by day and by source.
The real screens
Safe by design
Provider keys are stored as secrets and never returned by the platform. Model endpoints you register must be public and secure: private or internal addresses are refused so a tenant cannot point the platform at its own network.
Bring the model your policy allows.
Whether it is a hosted model or one in your own cloud account, it plugs in the same way.