Document Baseten integration

docs/user-guide/concepts/model-providers/baseten.md

@@ -0,0 +1,231 @@
+# Baseten
+
+[Baseten](https://baseten.co) is an AI inference provider that supports open-source, fine-tuned, and custom models of all modalities. The Strands Agents SDK can be used to run against powered by any OpenAI-compatible LLM hosted on Baseten.
+
+Baseten offers two options for LLM inference, both of which are compatible with the Strands Agents SDK:
+
+- **Model APIs**: Access to pre-deployed models like DeepSeek and Llama
+- **Dedicated Deployments**: Custom model deployments with dedicated infrastructure
+
+## Installation
+
+Baseten is configured as an optional dependency in Strands Agents. To install, run:
+
+```bash
+pip install 'strands-agents[baseten]'
+```
+
+## Configuration
+
+### API Key
+
+You'll need a Baseten API key to use the service. Set it as an environment variable:
+
+```bash
+export BASETEN_API_KEY="your-api-key-here"
+```
+
+### Model APIs
+
+Model APIs provide access to popular models through a common API endpoint. The SDK automatically uses the correct base URL for Model APIs:
+
+```python
+from strands.models.baseten import BasetenModel
+
+# DeepSeek R1 model
+model = BasetenModel(
+    model_id="deepseek-ai/DeepSeek-R1-0528",
+    client_args={
+        "api_key": os.getenv("BASETEN_API_KEY"),
+    },
+)
+
+# DeepSeek V3 model
+model = BasetenModel(
+    model_id="deepseek-ai/DeepSeek-V3-0324",
+    client_args={
+        "api_key": os.getenv("BASETEN_API_KEY"),
+    },
+)
+
+# Llama 4 Maverick model
+model = BasetenModel(
+    model_id="meta-llama/Llama-4-Maverick-17B-128E-Instruct",
+    client_args={
+        "api_key": os.getenv("BASETEN_API_KEY"),
+    },
+)
+
+# Llama 4 Scout model
+model = BasetenModel(
+    model_id="meta-llama/Llama-4-Scout-17B-16E-Instruct",
+    client_args={
+        "api_key": os.getenv("BASETEN_API_KEY"),
+    },
+)
+```
+
+Check the Baseten documentation for a [complete list of Model APIs](https://docs.baseten.co/development/model-apis/overview).
+
+### Dedicated Deployments
+
+Dedicated deployments provide custom model hosting with dedicated infrastructure. You'll need to specify the custom base URL:
+
+```python
+from strands.models.baseten import BasetenModel
+
+# Replace with your model URL
+base_url = "https://model-abcd1234.api.baseten.co/environments/production/sync/v1"
+
+model = BasetenModel(
+    model_id="",
+    base_url=base_url,
+    client_args={
+        "api_key": os.getenv("BASETEN_API_KEY"),
+    },
+)
+```
+
+### Model Parameters
+
+You can configure model parameters using the `params` argument:
+
+```python
+model = BasetenModel(
+    model_id="deepseek-ai/DeepSeek-V3-0324",
+    client_args={"api_key": os.getenv("BASETEN_API_KEY")},
+    params={
+        "max_tokens": 1000,
+        "temperature": 0.7
+    },
+)
+```
+
+## Usage
+
+After installing `strands-agents[baseten]`, you can import and initialize the Strands Agents' Baseten provider as follows:
+
+### Basic Usage
+
+```python
+from strands import Agent
+from strands.models.baseten import BasetenModel
+
+# Initialize model
+model = BasetenModel(
+    model_id="deepseek-ai/DeepSeek-V3-0324",
+    client_args={"api_key": os.getenv("BASETEN_API_KEY")},
+)
+
+# Create agent
+agent = Agent(model=model)
+
+# Chat with the model
+response = agent("Hello! How are you?")
+print(response)
+```
+
+### Async Usage
+
+The BasetenModel supports async operations:
+
+```python
+import asyncio
+from strands.models.baseten import BasetenModel
+
+async def chat_with_model():
+    model = BasetenModel(
+        model_id="deepseek-ai/DeepSeek-V3-0324",
+        client_args={"api_key": os.getenv("BASETEN_API_KEY")},
+    )
+
+    messages = [{"role": "user", "content": [{"text": "Hello!"}]}]
+
+    async for event in model.stream(messages):
+        print(event)
+
+# Run the async function
+asyncio.run(chat_with_model())
+```
+
+### Structured Output
+
+You can get structured output using Pydantic models:
+
+```python
+from pydantic import BaseModel
+from strands.models.baseten import BasetenModel
+
+class MathResult(BaseModel):
+    answer: int
+    explanation: str
+
+model = BasetenModel(
+    model_id="deepseek-ai/DeepSeek-V3-0324",
+    client_args={"api_key": os.getenv("BASETEN_API_KEY")},
+)
+
+messages = [{"role": "user", "content": [{"text": "What is 5 + 3?"}]}]
+
+async for result in model.structured_output(MathResult, messages):
+    print(f"Answer: {result['output'].answer}")
+    print(f"Explanation: {result['output'].explanation}")
+```
+
+### With Tools
+
+Baseten models support tool use for function calling:
+
+```python
+from strands.models.baseten import BasetenModel
+
+model = BasetenModel(
+    model_id="deepseek-ai/DeepSeek-V3-0324",
+    client_args={"api_key": os.getenv("BASETEN_API_KEY")},
+)
+
+tool_specs = [
+    {
+        "name": "calculator",
+        "description": "A simple calculator",
+        "inputSchema": {
+            "json": {
+                "type": "object",
+                "properties": {
+                    "expression": {"type": "string"}
+                },
+                "required": ["expression"]
+            }
+        }
+    }
+]
+
+messages = [{"role": "user", "content": [{"text": "Calculate 2 + 2"}]}]
+
+async for event in model.stream(messages, tool_specs=tool_specs):
+    print(event)
+```
+
+### Configuration Management
+
+You can update model configuration at runtime:
+
+```python
+model = BasetenModel(
+    model_id="deepseek-ai/DeepSeek-V3-0324",
+    client_args={"api_key": os.getenv("BASETEN_API_KEY")},
+    params={"max_tokens": 100}
+)
+
+# Update configuration
+model.update_config(params={"max_tokens": 200, "temperature": 0.8})
+
+# Get current configuration
+config = model.get_config()
+print(config)
+```
+
+## References
+
+* [Baseten Documentation](https://docs.baseten.co/)
+* [Strands SDK Documentation](https://docs.strands.ai/)

mkdocs.yml

@@ -85,6 +85,7 @@ nav:
       - Model Providers:
         - Amazon Bedrock: user-guide/concepts/model-providers/amazon-bedrock.md
         - Anthropic: user-guide/concepts/model-providers/anthropic.md
+        - Baseten: user-guide/concepts/model-providers/baseten.md
         - LiteLLM: user-guide/concepts/model-providers/litellm.md
         - LlamaAPI: user-guide/concepts/model-providers/llamaapi.md
         - MistralAI: user-guide/concepts/model-providers/mistral.md