Connect a model

The agent runs on your phone, but the AI model it talks to runs at a provider. Pipkin supports these:

Choice	What you need
Claude	A Claude Pro or Max subscription
OpenAI	An OpenAI Codex subscription
Gemini	A Gemini API key from Google AI Studio
Z.ai	An API key for the GLM Coding Plan
Kimi	An API key from the Kimi Code console
OpenAI-compatible API	A base URL and API key from any service that speaks the OpenAI API, including gateways and servers you run yourself

You can connect several and switch between them per chat. Add them during setup, or later in Settings → Providers → Add a provider.

Sign in with a subscription

1. Choose Claude or OpenAI.
2. Pick Browser login (default). Pipkin opens the sign-in page in your browser.
3. Sign in. When the browser sends you back, Pipkin shows Signed in.

If the browser doesn’t come back to Pipkin on its own, copy the code or the full address the browser ends up on, paste it into the sign-in sheet, and tap Continue.

The other options are useful when the browser flow doesn’t work on your phone:

• Copy code login (Claude): sign in, then paste the code Anthropic shows you.
• Device code login (OpenAI): Pipkin shows a code; open the link, enter the code, and Pipkin finishes on its own.

To sign out, open the subscription in Settings → Providers and tap Sign out.

Add a Gemini, Z.ai or Kimi key

1. Choose Gemini, Z.ai or Kimi.
2. Paste your API key and tap Continue. Pipkin shows Key saved.

Pipkin uses Pi’s built-in support for these providers, so their models, context windows and reasoning levels are already filled in. Usage counts against your Gemini API billing, your GLM Coding Plan, or your Kimi Code membership. Pipkin doesn’t check the key when you save it; a wrong key shows up as an error in the first chat.

Google doesn’t let third-party apps use a Google AI Pro or Ultra subscription, so Gemini needs an API key rather than a Google sign-in.

To remove a key, open the provider in Settings → Providers and tap Remove key.

Add an OpenAI-compatible API

1. In Settings → Providers, tap Add a provider, then OpenAI-compatible API.
2. Fill in:
   • Provider ID: a short name for it, such as deepinfra. Lowercase letters, digits, and dashes.
   • Base URL: the API’s address, which usually ends in /v1, such as https://api.example.com/v1.
   • API key: the key from your provider.
   • API type: Chat Completions works almost everywhere. Choose Responses if your provider supports OpenAI’s newer Responses API.
3. Tap Check. Pipkin asks the API for its list of models and shows Connected when it answers.
4. Tap Add models and pick from Available models, or type a Model ID under Custom models.
5. Tap Save.

Pipkin fills in what it knows about popular models. Tap a model to adjust its settings: whether it Accepts images, whether it Supports thinking and at which levels, its Context window and Max output tokens, and its cost per million tokens if you want Pipkin to show what a chat costs.

If the check fails, Pipkin explains why. The most common fixes:

• API key not accepted: check the key is complete and still active.
• No models list at this URL: the base URL usually needs to end in /v1.
• Not an OpenAI-compatible API: the URL points at a website rather than the API.
• Couldn’t reach the server or No response: check the address and that your phone is online.

To remove a provider, open it and tap Remove provider. Pipkin forgets the provider, its models, and its API key.

Choose a model for a chat

Under the message box, the model chip shows the current model. Tap it to search and switch. The Reasoning effort chip next to it sets how much the model thinks before answering; it only offers the levels the current model supports. You can also type /model or /thinking.

Whatever you pick applies to the current chat and becomes the default for new chats. To control which models show up in the list, open a provider in Settings → Providers and add or remove its models.

Where it’s stored

Pipkin saves providers in Pi’s own configuration files, in Termux under ~/.pipkin/pi:

• auth.json holds subscription sign-ins and the Gemini, Z.ai and Kimi keys.
• models.json holds OpenAI-compatible providers, their models, and their API keys.
• settings.json holds the default model and which models appear in the list.

These files contain your keys and sign-in tokens, so keep them private. A backup includes them.

Other providers

Pipkin’s screens cover the choices above, but Pi reads models.json directly, so you can add any provider Pi supports by editing the file. For example, a local Ollama server:

{
  "providers": {
    "ollama": {
      "baseUrl": "http://localhost:11434/v1",
      "api": "openai-completions",
      "apiKey": "ollama",
      "models": [{ "id": "qwen2.5-coder:7b" }]
    }
  }
}

Pipkin keeps anything it doesn’t manage when it saves the file. Pi’s models documentation covers every option. After editing, use Chat actions → Reload agent in an open chat to load the change.