Anthropic Provider¶

Integrates Anthropic's Claude models into Amplifier. Supports streaming, tool calling, extended thinking, and ChatRequest format.

Module ID¶

provider-anthropic

Installation¶

providers:
  - module: provider-anthropic
    source: git+https://github.com/microsoft/amplifier-module-provider-anthropic@main
    config:
      default_model: claude-sonnet-4-5

Configuration¶

Core Options¶

Option	Type	Default	Description
`api_key`	string	`$ANTHROPIC_API_KEY`	API authentication key
`base_url`	string	(none)	Custom API endpoint (for proxies, local APIs)
`default_model`	string	`claude-sonnet-4-5`	Default model
`max_tokens`	int	64000	Maximum response tokens
`temperature`	float	0.7	Sampling temperature
`timeout`	float	600.0	API timeout in seconds
`priority`	int	100	Provider priority for selection

Feature Toggles¶

Option	Type	Default	Description
`use_streaming`	bool	`true`	Use streaming API (required for large contexts)
`enable_prompt_caching`	bool	`true`	Enable prompt caching (90% cost reduction on cached tokens)
`enable_web_search`	bool	`false`	Enable native web search tool
`enable_1m_context`	bool	`false`	Enable 1M token context window (Sonnet and Opus only)
`filtered`	bool	`true`	Filter to curated model list

Rate Limit & Retry¶

Option	Type	Default	Description
`max_retries`	int	5	Total retry attempts before failing
`retry_jitter`	float	0.2	Randomness to add to delays (0.0-1.0)
`max_retry_delay`	float	60.0	Maximum wait between retries (seconds)
`min_retry_delay`	float	1.0	Minimum delay if no retry-after header
`throttle_threshold`	float	0.02	Pre-emptive throttle trigger (2% capacity remaining)
`throttle_delay`	float	1.0	Throttle delay when approaching limits (seconds)

Debug Options¶

Option	Type	Default	Description
`debug`	bool	`false`	Enable standard debug events
`raw_debug`	bool	`false`	Enable ultra-verbose raw API I/O logging

Supported Models¶

Model	Description
`claude-sonnet-4-5`	Claude Sonnet 4.5 (recommended, default)
`claude-opus-4-6`	Claude Opus 4.6 (most capable)
`claude-haiku-4-5`	Claude Haiku 4.5 (fastest, cheapest)

Extended Thinking¶

Enable extended thinking for complex reasoning tasks:

providers:
  - module: provider-anthropic
    config:
      default_model: claude-sonnet-4-5

Then in your request:

request = ChatRequest(
    model="claude-sonnet-4-5",
    messages=[...],
    enable_thinking=True
)

The response will include thinking blocks showing the model's reasoning process.

Prompt Caching¶

Anthropic's prompt caching reduces costs by 90% for repeated content. Enable it:

providers:
  - module: provider-anthropic
    config:
      enable_prompt_caching: true

The provider automatically marks eligible content for caching based on message structure.

Web Search¶

Enable native web search (Claude 4+ models only):

providers:
  - module: provider-anthropic
    config:
      enable_web_search: true

The model can then search the web directly during conversations.

Rate Limit Management¶

The provider tracks rate limits from response headers and pre-emptively throttles when approaching limits:

providers:
  - module: provider-anthropic
    config:
      throttle_threshold: 0.02  # Throttle at 2% capacity remaining
      throttle_safety_margin: 1.0  # Add 1s safety delay

Events: - provider:throttle - Emitted when pre-emptive throttling occurs - provider:retry - Emitted before each retry attempt

Retry Configuration¶

providers:
  - module: provider-anthropic
    config:
      max_retries: 5
      min_retry_delay: 1.0
      max_retry_delay: 60.0
      retry_jitter: 0.2

The provider uses exponential backoff with jitter and honors retry-after headers from the API.

Error Handling¶

The provider translates Anthropic SDK exceptions to kernel errors:

SDK Exception	Kernel Error	Retryable
`RateLimitError`	`RateLimitError`	Yes
`OverloadedError`	`ProviderUnavailableError`	Yes (10× backoff)
`InternalServerError`	`ProviderUnavailableError`	Yes
`AuthenticationError`	`AuthenticationError`	No
`BadRequestError` (context length)	`ContextLengthError`	No
`BadRequestError` (content filter)	`ContentFilterError`	No
`BadRequestError` (other)	`InvalidRequestError`	No

Debug Events¶

Enable debug logging to see request/response details:

providers:
  - module: provider-anthropic
    config:
      debug: true  # Standard debug events
      raw_debug: true  # Ultra-verbose raw API I/O

Events: - llm:request:debug - Request summary (message counts, model, params) - llm:response:debug - Response summary (usage, stop reason) - llm:request:raw - Complete unmodified request params - llm:response:raw - Complete unmodified response object

Environment Variables¶

export ANTHROPIC_API_KEY="your-api-key-here"

Example Configuration¶

providers:
  - module: provider-anthropic
    source: git+https://github.com/microsoft/amplifier-module-provider-anthropic@main
    config:
      api_key: ${ANTHROPIC_API_KEY}
      default_model: claude-sonnet-4-5
      max_tokens: 8192
      temperature: 1.0
      enable_prompt_caching: true
      enable_web_search: false
      max_retries: 5
      retry_jitter: 0.2