Models & providers
Every chat runs on a model, and most models are served by a provider you have connected. This page covers which providers are supported, where their keys are configured, how to use local models with Ollama, and how cost is tracked.
Choosing a model
Section titled “Choosing a model”Open a chat and use the ModelPicker to choose its model. The picker lists every model available for the providers you have configured (plus any local Ollama models). In a multi-model session, each chat can run on a different model so you can compare them on the same prompt.
Supported providers
Section titled “Supported providers”| Provider | Notes |
|---|---|
| OpenAI | Hosted models; requires an API key. |
| Gemini | Google’s models; requires an API key. |
| Mistral | Hosted models; requires an API key. |
| HuggingFace | Hosted inference; requires an API key. |
| NVIDIA Build | Hosted models; requires an API key. |
| Ollama | Local models running on your machine; no key required. |
Where keys are configured
Section titled “Where keys are configured”Provider API keys are not entered per chat — they live once in Settings and are reused everywhere (Chats, Studio, experiments). Add or update them under Providers & API keys.
Local models with Ollama
Section titled “Local models with Ollama”To run models locally — with no provider key and no data leaving your machine — install and run Ollama. Once it is reachable, its models appear in the ModelPicker alongside hosted ones. Ollama is also used for embeddings.
Cost and pricing
Section titled “Cost and pricing”Chatty keeps a pricing catalog that tracks the per-token cost of each model. This powers cost estimates and lets you treat cost as a first-class signal when comparing models or running experiments.