Providers & Models

Supported LLM providers, authentication methods, and model selection

Providers & Models

Pi supports multiple LLM providers through both API keys and OAuth subscriptions. Each provider offers different models with varying capabilities for vision, reasoning, and context windows.

Supported Providers

Pi maintains a registry of tool-capable models, updated with every release.

Subscription-Based (OAuth)

Anthropic Claude

Pro/Max subscription

pi
/login anthropic

OpenAI Codex

ChatGPT Plus/Pro subscription (GPT-5.x models)

pi
/login openai-codex

GitHub Copilot

Copilot subscription

pi
/login github-copilot

Google Gemini CLI

Gemini 2.0/2.5 via Cloud Code Assist

pi
/login google-gemini-cli

Google Antigravity

Free Gemini 3, Claude, GPT-OSS via Google Cloud

pi
/login google-antigravity

API Key-Based

Major Providers

Provider	Environment Variable	Models
Anthropic	ANTHROPIC_API_KEY	Claude 3.5 Sonnet, Opus, Haiku
OpenAI	OPENAI_API_KEY	GPT-4o, GPT-4o mini, o1
Google Gemini	GEMINI_API_KEY	Gemini 1.5/2.0 Pro, Flash
Google Vertex	GOOGLE_CLOUD_PROJECT + ADC	Gemini via Vertex AI

High-Performance

Provider	Environment Variable	Models
Groq	GROQ_API_KEY	Llama, Mixtral (fast inference)
Cerebras	CEREBRAS_API_KEY	Llama 3.3 (ultra-fast)
xAI	XAI_API_KEY	Grok 2, Grok Code

Aggregators & Enterprise

Provider	Environment Variable	Notes
OpenRouter	OPENROUTER_API_KEY	Access to 100+ models
Vercel AI Gateway	AI_GATEWAY_API_KEY	Multi-provider gateway
Azure OpenAI	AZURE_OPENAI_API_KEY + AZURE_OPENAI_BASE_URL	Enterprise OpenAI
Amazon Bedrock	AWS_ACCESS_KEY_ID + AWS_SECRET_ACCESS_KEY	Claude, Llama via AWS

Specialized

Provider	Environment Variable	Notes
Mistral	MISTRAL_API_KEY	Mistral Large, Codestral
MiniMax	MINIMAX_API_KEY	Chinese LLMs
Kimi For Coding	KIMI_API_KEY	Moonshot AI (Anthropic-compatible)
ZAI	ZAI_API_KEY	Specialized models

Authentication

Quick Start with OAuth

1. Start Pi

   pi
   
2. Login

   /login
   

   Select your provider from the interactive list.

3. Authenticate

   Follow the OAuth flow in your browser. Pi will automatically save your credentials to ~/.pi/agent/auth.json.

4. Select Model

   /model
   

   Choose from available models for your provider.

Using API Keys

Environment Variables

Set environment variables before starting Pi:

# Anthropic
export ANTHROPIC_API_KEY=sk-ant-...

# OpenAI
export OPENAI_API_KEY=sk-...

# Google Gemini
export GEMINI_API_KEY=AI...

# Then start Pi
pi

Command Line

Pass API key directly:

pi --api-key sk-ant-... --provider anthropic

Command-line API keys are visible in shell history. Use environment variables for better security.

.env File

Create .env in your project:

# .env
ANTHROPIC_API_KEY=sk-ant-...
OPENAI_API_KEY=sk-...

Load with:

source .env && pi

Or use a tool like direnv for automatic loading.

Special Authentication Cases

Google Vertex AI (ADC)

Vertex AI uses Application Default Credentials:

# Local development
gcloud auth application-default login
export GOOGLE_CLOUD_PROJECT="my-project"
export GOOGLE_CLOUD_LOCATION="us-central1"

# Production (service account)
export GOOGLE_APPLICATION_CREDENTIALS="/path/to/key.json"
export GOOGLE_CLOUD_PROJECT="my-project"
export GOOGLE_CLOUD_LOCATION="us-central1"

pi --provider google-vertex

See Google ADC documentation.

Azure OpenAI

Azure requires resource name or base URL:

export AZURE_OPENAI_API_KEY="your-key"
export AZURE_OPENAI_BASE_URL="https://your-resource.openai.azure.com/openai/v1"
# Or
export AZURE_OPENAI_RESOURCE_NAME="your-resource"

# Optional: API version (defaults to v1)
export AZURE_OPENAI_API_VERSION="v1"

# Optional: Model to deployment mapping
export AZURE_OPENAI_DEPLOYMENT_NAME_MAP="gpt-4o=my-deployment,gpt-4o-mini=prod"

pi --provider azure-openai

Amazon Bedrock

Bedrock uses AWS credentials:

# Option 1: AWS Profile
export AWS_PROFILE="my-profile"
export AWS_REGION="us-east-1"

# Option 2: Access Keys
export AWS_ACCESS_KEY_ID="..."
export AWS_SECRET_ACCESS_KEY="..."
export AWS_REGION="us-east-1"

# Option 3: Bearer Token
export AWS_BEARER_TOKEN_BEDROCK="..."
export AWS_REGION="us-east-1"

pi --provider amazon-bedrock

Model Selection

Interactive Selection

Press Ctrl+L or use /model to open the model selector:

┌─ Select Model ────────────────────────────────────┐
│ > claude-sonnet                                   │
│                                                   │
│ anthropic/claude-sonnet-4-20250514               │
│ anthropic/claude-opus-4-5                        │
│ anthropic/claude-3-5-haiku-20241022              │
│ openai/gpt-4o                                    │
│ openai/gpt-4o-mini                               │
│ google/gemini-2.5-flash                          │
└───────────────────────────────────────────────────┘

Command Line Selection

Basic

pi --provider anthropic --model claude-sonnet-4

# No --provider needed with prefix
pi --model anthropic/claude-sonnet-4

# Matches claude-sonnet-4-20250514
pi --model sonnet

# Shorthand syntax
pi --model sonnet:high

# Full syntax
pi --model sonnet --thinking high

Cycling Models

Configure models to cycle with Ctrl+P:

# Specific models
pi --models "claude-sonnet-4,claude-haiku,gpt-4o"

# All models from a provider
pi --models "anthropic/*"

# Glob patterns
pi --models "*sonnet*,*haiku*"

# With thinking levels
pi --models "sonnet:high,haiku:low,gpt-4o:medium"

In Pi, press:

• Ctrl+P: Next model
• Shift+Ctrl+P: Previous model

Listing Available Models

# List all models
pi --list-models

# Search models
pi --list-models sonnet
pi --list-models "vision"
pi --list-models "reasoning"

Model Capabilities

Vision Support

Models with input: ["text", "image"] can process images:

# Check if model supports vision
pi --list-models --model gpt-4o | grep input

# Use with images
pi --model gpt-4o @screenshot.png "What's in this image?"

Reasoning/Thinking

Models with reasoning: true support extended thinking:

OpenAI (o1, o3, gpt-5)

pi --model gpt-5-mini --thinking high

Thinking levels:

• minimal: Fast responses
• low, medium: Balanced
• high, xhigh: Deep reasoning

Anthropic (Claude Sonnet 4)

pi --model claude-sonnet-4 --thinking medium

Extended thinking blocks appear in responses with <thinking> tags.

Google (Gemini 2.5)

pi --model gemini-2.5-flash --thinking medium

Gemini shows reasoning process before final answers.

Default Models

Set default model in ~/.pi/agent/settings.json:

{
  "defaultProvider": "anthropic",
  "defaultModel": "claude-sonnet-4-20250514",
  "defaultThinkingLevel": "medium"
}

Or use /settings command in Pi.

Switching Providers Mid-Session

Pi supports switching models and providers during a conversation:

# In Pi:
> "Explain quantum computing" 
> (Using claude-sonnet-4)
> /model
> (Switch to gpt-4o)
> "Simplify that explanation"
> (Using gpt-4o, sees full conversation history)

Thinking blocks from one provider are converted to text with <thinking> tags when sent to other providers.

Cost & Token Tracking

Pi tracks token usage and costs automatically:

┌─ Footer ──────────────────────────────────────────┐
│ ~/project  session-name                           │
│ 1.2M / 200K tokens (cached: 800K)                │
│ $0.15 total ($0.03 this turn)                    │
│ anthropic/claude-sonnet-4                         │
└───────────────────────────────────────────────────┘

View detailed stats:

/session  # Shows total tokens, cost, cache usage

Environment Variables Reference

All Provider Variables

# Anthropic
ANTHROPIC_API_KEY=sk-ant-...
ANTHROPIC_OAUTH_TOKEN=...  # Alternative to API key

# OpenAI
OPENAI_API_KEY=sk-...

# Azure OpenAI
AZURE_OPENAI_API_KEY=...
AZURE_OPENAI_BASE_URL=https://resource.openai.azure.com/openai/v1
AZURE_OPENAI_RESOURCE_NAME=...  # Alternative to base URL
AZURE_OPENAI_API_VERSION=v1
AZURE_OPENAI_DEPLOYMENT_NAME_MAP=model=deployment,model2=deployment2

# Google
GEMINI_API_KEY=AI...
GOOGLE_CLOUD_PROJECT=my-project  # For Vertex/Gemini CLI
GOOGLE_CLOUD_LOCATION=us-central1
GOOGLE_APPLICATION_CREDENTIALS=/path/to/key.json

# Amazon Bedrock
AWS_PROFILE=my-profile
AWS_ACCESS_KEY_ID=...
AWS_SECRET_ACCESS_KEY=...
AWS_BEARER_TOKEN_BEDROCK=...
AWS_REGION=us-east-1

# Other Providers
GROQ_API_KEY=gsk_...
CEREBRAS_API_KEY=...
XAI_API_KEY=...
OPENROUTER_API_KEY=...
AI_GATEWAY_API_KEY=...
ZAI_API_KEY=...
MISTRAL_API_KEY=...
MINIMAX_API_KEY=...
KIMI_API_KEY=...

# GitHub Copilot
COPILOT_GITHUB_TOKEN=ghp_...
GH_TOKEN=ghp_...  # Alternative
GITHUB_TOKEN=ghp_...  # Alternative

# Advanced
PI_CACHE_RETENTION=long  # Extend cache: Anthropic 1h, OpenAI 24h
PI_AI_ANTIGRAVITY_VERSION=1.23.0  # Override Antigravity version

Troubleshooting

Authentication Failed

# Clear saved credentials
rm ~/.pi/agent/auth.json

# Re-authenticate
pi
/login

Model Not Available

# Check available models
pi --list-models

# Ensure you're authenticated
pi
/login provider-name

GitHub Copilot Model Not Supported

Enable the model in VS Code first:

1. Open VS Code
2. Open Copilot Chat
3. Click model selector
4. Select the model (shows warning icon)
5. Click “Enable”
6. Try again in Pi

Rate Limits

# Switch to different provider
/model

# Or use slower model from same provider
pi --model claude-haiku  # Cheaper, higher limits