Models and providers

PP Agent does not ship a model. You connect at least one provider, then choose a model for each conversation or agent.

Built-in providers

Open Settings › Models and sign in with an API key. Built-in providers include Anthropic, OpenAI, Google, OpenRouter, Mistral, Groq, and others supported by Pi.

Your key is stored encrypted in the app’s vault. The master key lives in your operating system’s keychain, not beside the encrypted data.

Custom gateways

If your team routes model traffic through its own gateway, add it under Settings › Models:

a base URL;
an API key;
the model names the gateway exposes;
whether it speaks the OpenAI or Anthropic API shape.

The gateway appears like any other provider. This is the usual setup for a company that wants central billing, logging, or an internal model.

Choosing a model

You can choose the model and the thinking level:

per conversation, from the model picker; and
per agent, so one agent always starts with the model that suits its work.

Thinking level controls how much reasoning the model may spend before answering. Higher levels help with analysis and planning; lower levels are faster and cheaper for extraction and rewriting.

Cost and tokens

The conversation shows token use and an estimated cost. Estimates depend on the prices the app knows for the provider; a custom gateway may not report prices, so treat the estimate as a guide.

Removing a provider

Remove a provider’s key in Settings › Models. Conversations that used it remain readable, but new turns need another model. Keys are never written to conversation files in plain text.