pi-kitBETA
GitHubDownload
Docs/Getting started

Connect a model

Provider authentication, LiteLLM, model selection, effort, and usage data.

Providers define how pi connects and authenticates. A thread selects a model from the available providers. Adding a connection does not necessarily change the default model or the model selected in an existing thread.

Add a provider#

Open Settings → Providers → +.
OptionConfiguration
RadiusSign in to pi's model gateway.
A subscriptionSelect a supported provider and complete its authorization flow.
An API keySelect the provider and enter its key.
A local modelConnect an OpenAI-compatible server.
LiteLLMConfigure a proxy URL and authentication, then discover its models.
Subscription support depends on the authentication methods available in the bundled pi version. API credentials and consumer subscriptions are separate options.
Add-provider options in Settings.
Add-provider options in Settings.

Sign in#

For Radius or a subscription:
  1. Select the provider.
  2. Complete browser authorization using the opened page or the fallback link/code.
  3. Return to pi-kit and wait for the connection state to update.
  4. Open the model picker and select a model.
For API-key providers, enter the key in the provider form. Review the default-model option before finishing. With an existing default, adding a provider does not automatically replace it.
The provider row reports credential storage. Keys entered through the supported workflow use the OS credential store when available. See Credential storage for resolver commands and fallback behavior.
OpenRouter is configured as an API-key provider. Its model availability, pricing, and limits come from the account associated with the key.

Connect LiteLLM#

Select LiteLLM in the provider wizard. Enter the proxy API URL, such as http://localhost:4000/v1, and choose an authentication mode:
  • A new API key, usually a proxy virtual key.
  • An existing saved key, when one is configured.
  • No key, if the proxy allows unauthenticated access.
Use Discover models to read the catalog available to that credential. Save and use models stores the connection. The model picker refreshes LiteLLM catalog data when opened.
Discovery reads model metadata; it does not run an inference request against each model. Context limits and capability flags are reported by the proxy.
LiteLLM connection form before credentials or a model catalog have been configured.
LiteLLM connection form before credentials or a model catalog have been configured.

Select a model#

Open the model control in the composer or header. Search the list and choose a model. Entries are grouped by provider, with pinned models listed separately.
  • Arrow keys move the highlighted entry; Enter selects it.
  • Shift+Enter pins or unpins the highlighted entry.
  • Model cycling uses the pinned set when configured.
  • Settings → Chat models contains Chat defaults, Pinned models, and Catalog tabs.
Changing the selected model retains the session history. The new model may have different context limits or tool support.

Set effort#

The brain button opens model-specific effort options. Available levels depend on the selected model. Defaults for new threads are separate from the current thread's selection.
Older builds placed effort inside the model picker; the current interface uses a separate popover.

Usage and context#

The interface can display several different measurements:
  • Context: the session's use of the model's input budget.
  • Provider usage: account limits or credits returned by a supported adapter.
  • Cost: usage-derived estimates rather than an invoice.
Not every provider supplies usage data. Local servers generally show their address. OpenRouter may return no balance for unrestricted inference keys, and the documented adapter does not supply OpenAI subscription usage. Missing data is left blank.