> ## Documentation Index
> Fetch the complete documentation index at: https://daily-docs-pr-5482.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.
# Anthropic LLM
> AnthropicLLMService integrates Claude models with streaming responses, function calling, and prompt caching support.
## Overview
`AnthropicLLMService` provides integration with Anthropic's Claude models, supporting streaming responses, function calling, and prompt caching with specialized context handling for Anthropic's message format and advanced reasoning capabilities.
Pipecat's API methods for Anthropic Claude integration
Complete example with function calling
Official Anthropic API documentation and features
Access Claude models and API keys
## Installation
To use Anthropic services, install the required dependency:
```bash theme={null}
uv add "pipecat-ai[anthropic]"
```
## Prerequisites
### Anthropic Account Setup
Before using Anthropic LLM services, you need:
1. **Anthropic Account**: Sign up at [Anthropic Console](https://console.anthropic.com/)
2. **API Key**: Generate an API key from your console dashboard
3. **Model Selection**: Choose from available Claude models (Claude Sonnet 4.5, Claude Opus 4.6, etc.)
### Required Environment Variables
* `ANTHROPIC_API_KEY`: Your Anthropic API key for authentication
## Configuration
Anthropic API key for authentication.
Claude model name to use (e.g., `"claude-sonnet-4-5-20250929"`,
`"claude-opus-4-6-20250929"`). *Deprecated in v0.0.105. Use
`settings=AnthropicLLMService.Settings(...)` instead.*
Runtime-configurable model settings. See [Settings](#settings) below.
Runtime-configurable model settings. See [Settings](#settings) below.
*Deprecated in v0.0.105. Use `settings=AnthropicLLMService.Settings(...)`
instead.*
Optional custom Anthropic client instance. Useful for custom clients like
`AsyncAnthropicBedrock` or `AsyncAnthropicVertex`.
Request timeout in seconds. Used when `retry_on_timeout` is enabled to
determine when to retry.
Whether to retry the request once if it times out. The retry attempt has no
timeout limit.
### Settings
Runtime-configurable settings passed via the `settings` constructor argument using `AnthropicLLMService.Settings(...)`. These can be updated mid-conversation with `LLMUpdateSettingsFrame`. See [Service Settings](/pipecat/fundamentals/service-settings) for details.
| Parameter | Type | Default | Description |
| ----------------------- | ------------------------- | ----------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `model` | `str` | `None` | Anthropic model identifier. *(Inherited from base settings.)* |
| `system_instruction` | `str` | `None` | System instruction/prompt for the model. *(Inherited from base settings.)* |
| `max_tokens` | `int` | `NOT_GIVEN` | Maximum tokens to generate. |
| `temperature` | `float` | `NOT_GIVEN` | Sampling temperature (0.0 to 1.0). Lower values are more focused, higher values more creative. |
| `top_k` | `int` | `NOT_GIVEN` | Top-k sampling parameter. Limits tokens to the top k most likely. |
| `top_p` | `float` | `NOT_GIVEN` | Top-p (nucleus) sampling (0.0 to 1.0). Controls diversity of output. |
| `enable_prompt_caching` | `bool` | `NOT_GIVEN` | Whether to enable Anthropic's prompt caching feature. Reduces costs for repeated context. |
| `thinking` | `AnthropicThinkingConfig` | `NOT_GIVEN` | Thinking configuration. If not provided, Pipecat disables thinking on Sonnet 5 and later, which otherwise decide per request whether to think, to reduce latency; Opus and Fable are left at Anthropic's default. See [AnthropicThinkingConfig](#anthropicthinkingconfig) below. |
`NOT_GIVEN` values are omitted from the API request entirely, letting the
Anthropic API use its own defaults.
### AnthropicThinkingConfig
Configuration for Anthropic's thinking feature, which causes the model to spend more time reasoning before responding.
| Parameter | Type | Default | Description |
| --------------- | ------------------------------------------ | ------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `type` | `"adaptive"`, `"enabled"`, or `"disabled"` | | Thinking mode. `"adaptive"` lets the model decide when and how deeply to think; prefer it. `"enabled"` is legacy manual thinking, sized by `budget_tokens`: Claude 4.7 and later reject it, and Claude 4.5 and earlier accept only it. `"disabled"` turns thinking off. |
| `budget_tokens` | `int` (optional) | `None` | Maximum number of tokens for thinking. With today's models, the minimum is 1024. Required when type is `"enabled"`, not allowed otherwise. |
| `display` | `"summarized"` or `"omitted"` (optional) | `None` | How thinking text comes back: `"summarized"` for readable thinking, which is what `LLMThoughtTextFrame` carries, or `"omitted"` for thinking blocks whose text is empty. Claude 4.7 and later default to `"omitted"`, so set `"summarized"` there to keep those frames carrying text. Not allowed when type is `"disabled"`. |
When thinking is enabled, the service emits `LLMThoughtStartFrame`, `LLMThoughtTextFrame`, and `LLMThoughtEndFrame` during response generation.
## Usage
### Basic Setup
```python theme={null}
from pipecat.services.anthropic.llm import AnthropicLLMService
llm = AnthropicLLMService(
api_key=os.getenv("ANTHROPIC_API_KEY"),
model="claude-sonnet-4-5-20250929",
)
```
### With Custom Settings
```python theme={null}
from pipecat.services.anthropic.llm import AnthropicLLMService
llm = AnthropicLLMService(
api_key=os.getenv("ANTHROPIC_API_KEY"),
settings=AnthropicLLMService.Settings(
model="claude-sonnet-4-5-20250929",
enable_prompt_caching=True,
max_tokens=2048,
temperature=0.7,
),
)
```
### With Thinking
```python theme={null}
# Adaptive thinking (preferred for Claude 4.7+)
llm = AnthropicLLMService(
api_key=os.getenv("ANTHROPIC_API_KEY"),
settings=AnthropicLLMService.Settings(
model="claude-sonnet-4-7-20250929",
max_tokens=16384,
thinking=AnthropicLLMService.AnthropicThinkingConfig(
type="adaptive",
display="summarized", # Get readable thinking text
),
),
)
# Legacy manual thinking (for Claude 4.5 and earlier)
llm = AnthropicLLMService(
api_key=os.getenv("ANTHROPIC_API_KEY"),
settings=AnthropicLLMService.Settings(
model="claude-sonnet-4-5-20250929",
max_tokens=16384,
thinking=AnthropicLLMService.AnthropicThinkingConfig(
type="enabled",
budget_tokens=10000,
),
),
)
```
### Updating Settings at Runtime
Model settings can be changed mid-conversation using `LLMUpdateSettingsFrame`:
```python theme={null}
from pipecat.frames.frames import LLMUpdateSettingsFrame
from pipecat.services.anthropic.llm import AnthropicLLMSettings
await worker.queue_frame(
LLMUpdateSettingsFrame(
delta=AnthropicLLMSettings(
temperature=0.3,
max_tokens=1024,
)
)
)
```
The `InputParams` / `params=` pattern is deprecated as of v0.0.105. Use
`Settings` / `settings=` instead. See the [Service Settings
guide](/pipecat/fundamentals/service-settings) for migration details.
## Notes
* **Prompt caching**: When `enable_prompt_caching` is enabled, Anthropic caches repeated context to reduce costs. Cache control markers are automatically added to the most recent user messages. This is most effective for conversations with large system prompts or long conversation histories.
* **Thinking**: Enabling thinking increases response quality for complex tasks but adds latency. Use `type="adaptive"` (preferred for Claude 4.7+) to let the model decide when to think, or `type="enabled"` (legacy mode for Claude 4.5 and earlier, requires `budget_tokens` with minimum 1024). Claude 4.7+ models default to `display="omitted"`, so set `display="summarized"` to receive readable thinking text in `LLMThoughtTextFrame`. When the `thinking` parameter is not set, Pipecat disables thinking on Sonnet 5 and later (which otherwise decide per request whether to think) to keep latency low for real-time voice; Opus and Fable are left at Anthropic's default.
* **Prompt-elicited `` tags**: If your system prompt asks the model to reason inside inline tags rather than enabling thinking via the `thinking` parameter, that reasoning is ordinary text and will be spoken by TTS. Prefer the `thinking` parameter; for inline tags you deliberately keep, see [Removing Tagged Content](/api-reference/server/utilities/text/pattern-pair-aggregator#removing-tagged-content).
* **Custom clients**: You can pass custom Anthropic client instances (e.g., `AsyncAnthropicBedrock` or `AsyncAnthropicVertex`) via the `client` parameter to use Anthropic models through other cloud providers.
* **Retry behavior**: When `retry_on_timeout=True`, the first attempt uses the `retry_timeout_secs` timeout. If it times out, a second attempt is made with no timeout limit.
* **System instruction precedence**: If both `system_instruction` (from the constructor) and a system message in the context are set, the constructor's `system_instruction` takes precedence and a warning is logged.
## Event Handlers
`AnthropicLLMService` supports the following event handlers, inherited from [LLMService](/api-reference/server/events/service-events):
| Event | Description |
| --------------------------- | ----------------------------------------------------------------------- |
| `on_completion_timeout` | Called when an LLM completion request times out |
| `on_function_calls_started` | Called when function calls are received and execution is about to start |
```python theme={null}
@llm.event_handler("on_completion_timeout")
async def on_completion_timeout(service):
print("LLM completion timed out")
```