LLM Providers

TAU supports 75+ LLM providers. Use your API keys or connect existing CLI tools for free.

Use Your Existing Subscription

Connect to Claude Code, GitHub Copilot, Gemini CLI, or Codex — no extra API costs:

# Claude Code (Claude Pro/Max subscription)
tau auth login claude-code

# GitHub Copilot (GitHub subscription)
tau auth login github-copilot

# Gemini CLI (Google AI subscription)
tau auth login gemini-cli

# Codex CLI
tau auth login codex

TAU acts as an ACP (Agent Client Protocol) adapter — you use models through existing subscriptions.

Quick Setup (API Keys)

# Set API key for your preferred provider
export ANTHROPIC_API_KEY=sk-ant-xxx    # Claude
export OPENAI_API_KEY=sk-xxx           # GPT
export GOOGLE_API_KEY=xxx              # Gemini
export XAI_API_KEY=xxx                 # Grok
export DEEPSEEK_API_KEY=xxx            # DeepSeek

# Run TAU - it auto-detects configured providers
tau

Tier 1: Native Providers

These have dedicated implementations with full feature support:

Provider	Env Variable	Models	Features
Anthropic	`ANTHROPIC_API_KEY`	Claude Opus 4.5, Sonnet 4.5, Haiku 4.5	Streaming, tools, vision, prompt caching
OpenAI	`OPENAI_API_KEY`	GPT-5.2, GPT-4o, o1, o3	Streaming, tools, vision, function calling
Google	`GOOGLE_API_KEY`	Gemini 3 Pro, 3 Flash	Streaming, tools, vision, context caching
xAI	`XAI_API_KEY`	Grok 4.1, Grok Code Fast 1	Streaming, tools, fast inference
AWS Bedrock	`AWS_ACCESS_KEY_ID`	Claude, Nova, Llama	Cross-region inference
OpenRouter	`OPENROUTER_API_KEY`	300+ models	Gateway to all providers
Ollama	`OLLAMA_HOST`	Llama 4, Mistral, Qwen, etc.	Local, free, offline
MLX	(auto-detected)	Local embeddings	macOS Apple Silicon only

CLI Providers (Use Your Subscription)

Connect to existing CLI agents and use models through your subscription:

Claude Code (Claude Pro/Max)

# Requires: Claude Pro/Max subscription + Claude Code CLI installed
# Install: npm install -g @anthropic/claude-code

# Authenticate
tau auth login claude-code

# Use Claude Code models (free with subscription)
tau run --provider claude-code "Explain this codebase"

# Or set as default
export TAU_ACP_AGENT=claude

GitHub Copilot

# Requires: GitHub Copilot subscription

# Authenticate via OAuth
tau auth login github-copilot

# Use Copilot models (free with subscription)
tau run --model github-copilot/gpt-4o "Hello"
tau run --model github-copilot/claude-4-5-sonnet "Hello"

# Available models depend on your subscription tier

Gemini CLI

# Requires: Gemini CLI installed + Google AI subscription

# Authenticate
tau auth login gemini-cli

# Use Gemini
tau run --provider gemini-cli "Explain this code"

export TAU_ACP_AGENT=gemini

Codex CLI

# Requires: Codex CLI installed

# Authenticate
tau auth login codex

# Use Codex
tau run --provider codex "Refactor this function"

export TAU_ACP_AGENT=codex

Note: CLI providers use ACP (Agent Client Protocol). TAU acts as an adapter — the actual inference happens through the external CLI, not through direct API calls. This means your subscription limits apply.

Tier 2: models.dev Providers (70+)

All providers from models.dev are supported via OpenAI-compatible API:

xAI

Grok Code Fast 1

Mistral

Large, Medium

Groq

Llama 4 Scout

DeepSeek

V3.2

Cerebras

Llama 3.3

Cohere

Command-R+

Together AI

Open models

Perplexity

Sonar

Fireworks

Fast inference

MiniMax

M2.1

Zhipu AI

GLM-4.7

Alibaba

Qwen 2.5

+ 60 more providers. Run curl -s https://models.dev/api.json | jq 'keys' to see all.

Switching Providers

# Use specific provider/model
tau run --model anthropic/claude-sonnet-4-5-20250929 "Hello"
tau run --model openai/gpt-4o "Hello"
tau run --model xai/grok-code-fast-1 "Hello"

# Set default model
export TAU_MODEL=anthropic/claude-sonnet-4-5-20250929

# List configured providers
tau models list

# List models from specific provider
tau models list --provider anthropic

Local Models (Ollama)

# Start Ollama
ollama serve

# Pull a model
ollama pull llama3.2:3b

# Configure TAU to use Ollama
export OLLAMA_HOST=http://localhost:11434
export TAU_ENABLE_OLLAMA=1

# Run with Ollama
tau run --model ollama/llama3.2:3b "Hello"

Enterprise Providers

AWS Bedrock

export AWS_ACCESS_KEY_ID=AKIA...
export AWS_SECRET_ACCESS_KEY=...
export AWS_REGION=us-east-1

tau run --model amazon-bedrock/anthropic.claude-sonnet-4-5-20250929-v1:0 "Hello"

Azure OpenAI

export AZURE_OPENAI_API_KEY=xxx
export AZURE_OPENAI_ENDPOINT=https://my-resource.openai.azure.com

tau run --model azure-openai/my-gpt4-deployment "Hello"

Google Vertex AI

# Authenticate with GCP
gcloud auth application-default login

export GOOGLE_CLOUD_PROJECT=my-project
export VERTEX_LOCATION=us-central1

tau run --model google-vertex/gemini-3-pro "Hello"

Cost Tracking

TAU tracks costs for all providers using models.dev pricing data:

# See cost breakdown after conversation
tau history --costs

# Real-time cost shown in TUI status bar

# CLI providers show $0 (uses your subscription)