Friends of PP

For the few who know, the pool awaits. Unlimited resources. Zero friction.

AI Pool

Choose your setup path

Start with the CLI you actually use. Each panel gives you the automatic installer first, then the raw files if you want to inspect or paste by hand.

CLIs

Codex

Claude

Gemini

Cute Code

SDKs

Anthropic API

OpenAI API

Pool

Analytics

Step 1: Install CLI

# Using NPM (All Platforms)
npm install -g @openai/codex

# Using Homebrew (macOS)
brew install codex

Automatic Config

Manual Config

One-Liner Execution

macOS / Linux (bash)

Windows (PowerShell)

Automatically retrieves auth token and configures local environment.

01 // Auth Credentials

Target: ~/.codex/auth.json

02 // Configuration

Target: ~/.codex/config.toml

Step 1: Install CLI

# Using NPM (All Platforms)
npm install -g @google/gemini-cli

# Using Homebrew (macOS)
brew install gemini-cli

Automatic Config

Manual Config

One-Liner Execution

macOS / Linux (bash)

Windows (PowerShell)

Automatically retrieves OAuth token and configures endpoint settings.

01 // Configuration

Target: ~/.gemini/settings.json + environment variable

macOS / Linux (bash/zsh)

Windows (PowerShell)

02 // Settings Only

Target: ~/.gemini/settings.json

Step 1: Install CLI

# Using NPM (All Platforms)
npm install -g @anthropic-ai/claude-code

# Using Homebrew (macOS)
brew install claude-code

Automatic Config

Manual Config

One-Liner Execution

macOS / Linux (bash/zsh)

Windows (PowerShell)

Sets ANTHROPIC_BASE_URL, CLAUDE_CODE_OAUTH_TOKEN, and skips onboarding.

01 // Shell Environment (Required)

Add to ~/.bashrc or ~/.zshrc (macOS/Linux) or $PROFILE (PowerShell). Claude Code reads env vars for auth.

macOS / Linux (bash/zsh)

Windows (PowerShell)

02 // Skip Onboarding

Add to ~/.claude.json

{"hasCompletedOnboarding": true}

03 // Available Models

Use these model names in Claude Code (routed through the pool)

haiku sonnet opus k2p5 kimi-k2-thinking MiniMax-M2.7 MiniMax-M2.7-highspeed

04 // Optional: Settings.json

Claude Code ignores auth in ~/.claude/settings.json; this is optional for other settings

Anthropic-Compatible API

Use the pool token as an Anthropic API key. Official Anthropic clients send it as x-api-key, and the pool routes Claude models natively while translating GPT/Kimi/MiniMax/GLM models through the same /v1/messages interface.

Base URL

Auth

ANTHROPIC_API_KEY can be the pool token.

Models

Claude models plus GPT/Kimi/MiniMax/GLM via translation.

Anthropic SDK

Claude Agent SDK

Env + curl

01 // Install

pip install anthropic

02 // Messages API

This is the normal Anthropic Python SDK. The only difference is base_url and the pool token.

01 // Install

npm install @anthropic-ai/claude-agent-sdk

02 // Claude Agent SDK

Use this when you want Claude Code's agent loop programmatically. It reads ANTHROPIC_BASE_URL and CLAUDE_CODE_OAUTH_TOKEN.

01 // Environment

02 // curl

Model Names

claude-sonnet-4-6 claude-opus-4-7 gpt-5.5 gpt-5.4 k2p5 MiniMax-M2.7 glm-5.1

OpenAI-Compatible API

Use the pool with the official OpenAI SDK, the OpenAI Agents SDK, Cursor, Continue, Aider, LiteLLM, or any OpenAI-compatible client. Chat Completions and Responses both work; model names route to the right provider automatically.

Base URL

Auth

OPENAI_API_KEY is the same pool token.

Interfaces

chat.completions, responses, and Agents SDK models.

01 // Environment Variables

Set these in your shell or tool config.

macOS / Linux (bash/zsh)

Windows (PowerShell)

OpenAI SDK

Python Agents

TypeScript Agents

Copy Paste

curl

02 // Install

pip install openai

03 // Chat Completions

04 // Legacy Completions

Use this for older packages that still call /v1/completions. The pool translates it to Codex Responses and returns normal text_completion JSON.

05 // Responses API

The proxy normalizes the official SDK's Responses payload into the Codex backend shape and returns normal non-streaming Responses JSON.

06 // Images API

images.generate supports n by fanning out through the pool. images.edit accepts the official SDK multipart upload and translates it to the Responses image_generation tool with an input image.

02 // Install

pip install openai-agents

03 // OpenAI Agents SDK for Python

Use OpenAIResponsesModel for the normal Agents path. OpenAIChatCompletionsModel also works if a project is built around chat completions.

02 // Install

npm install @openai/agents openai zod

03 // OpenAI Agents SDK for TypeScript

This follows the current OpenAI Agents SDK shape: create an OpenAI client with the pool base URL, set it as the default client, then use Responses or Chat Completions models.

To integrate, copy paste this in to your agent:

This gives another agent or code assistant the exact base URL, auth header, model names, and endpoint choices it needs to swap an OpenAI client onto the pool.

02 // curl

Model Names

Use any model name. Requests are auto-routed to the right provider and translated if needed.

gpt-5.5 gpt-5.4 gpt-5.3-codex claude-sonnet-4-6 claude-opus-4-7 k2p5 MiniMax-M2.7 glm-5.1

Claude models sent to /v1/chat/completions are translated to Anthropic Messages and back. OpenAI models sent to /v1/messages are translated to Codex Responses and back.

Pi Setup

Pi gets its own models.json. Codex runs through openai-codex-responses, while Claude, Kimi, and MiniMax run through anthropic-messages.

01 // Install models.json

Target: ~/.pi/agent/models.json

macOS / Linux (bash/zsh)

Windows (PowerShell)

02 // Curated Providers

Pi only gets the pool-backed providers, so the built-in Anthropic clutter stays out of the picker.

claude kimi minimax codex zai

claude-haiku-4-5 claude-sonnet-4-6 claude-opus-4-7 claude-opus-4-6 k2p5 kimi-k2-thinking MiniMax-M2.7 MiniMax-M2.7-highspeed glm-5.1

03 // Raw models.json

Replace the file, then fully restart Pi so it reloads the provider list.

Cute Code Setup

Installs cute-code, writes pool settings to ~/.claude/settings.json, and registers Codex/OpenAI, Claude, Kimi, MiniMax, and GLM models.

01 // Install and configure

macOS / Linux (bash/zsh)

Windows (PowerShell)

02 // Generated settings.json

Target: ~/.claude/settings.json

After setup, run cute-code --model gpt-5.5 or pick any configured model in the model selector.

Pool Analytics

Connecting to matrix...

Terminate Session