Provider Setup¶

Amplifier supports multiple LLM providers. This guide covers setting up each one.

Supported Providers¶

Provider	Models	Best For
Anthropic	Claude 4 Sonnet, Opus, Haiku	General purpose, coding, analysis
OpenAI	GPT-5.4, GPT-5-mini, GPT-5-nano	Broad capabilities, function calling
Azure OpenAI	GPT-5.4 via Azure	Enterprise, compliance requirements
Ollama	Llama, Mistral, etc.	Local/offline, privacy, experimentation
vLLM	Any vLLM-compatible	Self-hosted inference, high throughput

For detailed configuration options and advanced features, see Provider Modules.

Anthropic (Recommended)¶

Get an API Key¶

Visit console.anthropic.com
Create an account or sign in
Navigate to API Keys
Create a new API key

Configure¶

# Set environment variable
export ANTHROPIC_API_KEY="sk-ant-..."

# Or add to shell profile
echo 'export ANTHROPIC_API_KEY="sk-ant-..."' >> ~/.bashrc

# Select as provider
amplifier provider use anthropic

Available Models¶

Model	Description
`claude-sonnet-4-5`	Recommended, balanced performance (default)
`claude-opus-4-6`	Most capable, best for complex tasks
`claude-haiku-4-5`	Fastest, cheapest

# Use a specific model
amplifier run --model claude-opus-4-6 "Complex analysis task"

Advanced Configuration¶

Debug Logging:

providers:
  - module: provider-anthropic
    config:
      debug: true      # Enable standard debug events
      raw_debug: true  # Enable ultra-verbose raw API I/O logging

Retry Configuration:

providers:
  - module: provider-anthropic
    config:
      max_retries: 5              # Maximum retry attempts (default: 5)
      min_retry_delay: 1.0        # Base delay in seconds (default: 1.0)
      max_retry_delay: 60.0       # Cap on base delay (default: 60.0)
      retry_jitter: 0.2           # Jitter fraction 0.0-1.0 (default: 0.2)
      overloaded_delay_multiplier: 10.0  # Multiplier for 529 errors (default: 10.0)

Beta Headers (1M Token Context):

providers:
  - module: provider-anthropic
    config:
      default_model: claude-sonnet-4-5
      beta_headers: "context-1m-2025-08-07"  # Enable 1M token context window
      max_tokens: 8192  # Output tokens remain separate from context window

OpenAI¶

Get an API Key¶

Visit platform.openai.com
Create an account or sign in
Navigate to API Keys
Create a new API key

Configure¶

# Set environment variable
export OPENAI_API_KEY="sk-..."

# Or add to shell profile
echo 'export OPENAI_API_KEY="sk-..."' >> ~/.bashrc

# Select as provider
amplifier provider use openai

Available Models¶

Model	Description
`gpt-5.4`	Optimized for code, recommended (default)
`gpt-5-mini`	Smaller, faster GPT-5
`gpt-5-nano`	Smallest GPT-5 variant

# Use a specific model
amplifier run --model gpt-5.4 "Complex analysis task"

Advanced Configuration¶

Debug Logging:

providers:
  - module: provider-openai
    config:
      debug: true      # Enable standard debug events
      raw_debug: true  # Enable ultra-verbose raw API I/O logging

Reasoning Control:

providers:
  - module: provider-openai
    config:
      reasoning: "low"              # Reasoning effort: minimal|low|medium|high
      reasoning_summary: "detailed" # Reasoning verbosity: auto|concise|detailed

Automatic Context Management:

providers:
  - module: provider-openai
    config:
      truncation: "auto"  # Automatic context management (default: "auto")
      # OR
      truncation: null    # Disables automatic truncation (manual control)

Azure OpenAI¶

Prerequisites¶

Azure subscription
Azure OpenAI resource created
Model deployment created

Configure¶

With API Key:

# Set environment variables
export AZURE_OPENAI_ENDPOINT="https://myresource.openai.azure.com"
export AZURE_OPENAI_API_KEY="your-api-key-here"
export AZURE_OPENAI_DEFAULT_MODEL="gpt-5.4"

# Select as provider
amplifier provider use azure-openai

With DefaultAzureCredential (Recommended):

# Login to Azure
az login

# Set environment variables
export AZURE_OPENAI_ENDPOINT="https://myresource.openai.azure.com"
export AZURE_USE_DEFAULT_CREDENTIAL="true"
export AZURE_OPENAI_DEFAULT_MODEL="gpt-5.4"

# Select as provider
amplifier provider use azure-openai

Deployment Mapping¶

Azure uses deployment names instead of model names:

providers:
  - module: provider-azure-openai
    config:
      azure_endpoint: "https://myresource.openai.azure.com"
      api_key: "${AZURE_OPENAI_API_KEY}"
      deployment_mapping:
        "gpt-5.4": "my-gpt5-deployment"
        "gpt-5-mini": "my-mini-deployment"
      default_deployment: "fallback-deployment"

Advanced Configuration¶

Debug Logging:

providers:
  - module: provider-azure-openai
    config:
      debug: true      # Enable standard debug events
      raw_debug: true  # Enable ultra-verbose raw API I/O logging

Managed Identity:

providers:
  - module: provider-azure-openai
    config:
      azure_endpoint: "https://myresource.openai.azure.com"
      use_managed_identity: true
      # Optional: for user-assigned managed identity
      managed_identity_client_id: "client-id-here"

Ollama (Local Models)¶

Prerequisites¶

Install Ollama: https://ollama.ai
Pull a model: ollama pull llama3.2:3b
Start Ollama server (usually automatic)

Configure¶

# Set environment variable (optional, defaults to localhost)
export OLLAMA_HOST="http://localhost:11434"

# Select as provider
amplifier provider use ollama

Available Models¶

Any model from https://ollama.ai/library:

Model	Description
`llama3.2:3b`	Small, fast (recommended)
`llama3.2:1b`	Tiny, fastest
`mistral`	7B general purpose
`codellama`	Code generation
`deepseek-r1`	Reasoning/thinking
`qwen3`	Reasoning + tools

# Use a specific model
amplifier run --model llama3.2:3b "Explain this code"

Advanced Configuration¶

Debug Logging:

providers:
  - module: provider-ollama
    config:
      debug: true      # Enable standard debug events
      raw_debug: true  # Enable ultra-verbose raw API I/O logging

Auto-pull Models:

providers:
  - module: provider-ollama
    config:
      host: "http://localhost:11434"
      default_model: "llama3.2:3b"
      auto_pull: true  # Automatically pull missing models
      timeout: 300     # Request timeout in seconds (default: 5 minutes)

Thinking/Reasoning:

providers:
  - module: provider-ollama
    config:
      default_model: "deepseek-r1"  # Or qwen3, qwq, phi4-reasoning
      # Enable thinking in ChatRequest with enable_thinking=True

Troubleshooting¶

Common Issues¶

Anthropic: - Invalid API key → Check ANTHROPIC_API_KEY format - Rate limits → Adjust retry configuration - 529 Overloaded → Provider retries automatically with 10x backoff

OpenAI: - Invalid API key → Check OPENAI_API_KEY format - Incomplete responses → Provider auto-continues up to 5 attempts - Context limits → Enable truncation: "auto" for automatic management

Azure OpenAI: - Authentication errors → Verify endpoint URL and credentials - Deployment not found → Check deployment name mapping - WSL2 + Managed Identity → Use DefaultAzureCredential instead

Ollama: - Server offline → Check ollama serve is running - Model not found → Run ollama pull <model> or enable auto_pull - Connection issues → Verify OLLAMA_HOST or host config

Debug Mode¶

Enable debug logging to see detailed request/response information:

providers:
  - module: provider-anthropic  # or provider-openai, provider-azure-openai, provider-ollama
    config:
      debug: true      # Standard debug events
      raw_debug: true  # Ultra-verbose raw API I/O (use sparingly)

Next Steps¶

Configure Tools - Add capabilities to your agent
Session Management - Manage conversation history
Profiles - Create reusable configurations