models: Add support for DeepSeek models

[veeragoni]

Features

• New DeepSeek Model Provider: Full implementation with streaming and tool use support
• OpenAI-compatible API: Uses OpenAI client for seamless integration with DeepSeek endpoints
• Tool Use Support: Proper tool calling with calculator, file operations, and shell commands
• Reasoning Model Support: Handles DeepSeek-specific features like reasoning content
• Beta Features: Support for beta endpoint and advanced DeepSeek capabilities

Changes

• src/strands/models/deepseek.py: New DeepSeek model provider implementation with proper tool use streaming
• src/strands/models/__init__.py: Export DeepSeekModel class
• tests_integ/models/test_model_deepseek.py: Comprehensive integration tests (8 test cases)
• tests_integ/models/ds_test.py: Demonstration script showing tool use capabilities
• README.md: Update documentation with DeepSeek examples and provider list

Usage

from strands.models.deepseek import DeepSeekModel
from strands_tools import calculator

model = DeepSeekModel(api_key="your-key", model_id="deepseek-chat")
agent = Agent(model=model, tools=[calculator])
result = agent("What is 42 ^ 9")  # Uses calculator tool correctly

Testing

✅ All 8 integration tests passing
✅ Basic conversation, structured output, streaming, tool usage
✅ Multi-tool workflows (calculator, file_read, shell)
✅ Configuration updates and async operations
✅ Tool use streaming matches OpenAI/Ollama implementations
✅ All mypy type checks passing

Documentation PR

strands-agents/docs#135

Type of Change

New feature

How have you tested the change? Verify that the changes do not break functionality or introduce warnings in consuming repositories: agents-docs, agents-tools, agents-cli

I ran hatch run prepare

Checklist

[x] I have read the CONTRIBUTING document
[x] I have added any necessary tests that prove my fix is effective or my feature works
[x] I have updated the documentation accordingly
[x] I have added an appropriate example to the documentation to outline the feature, or no new docs are needed
[x] My changes generate no new warnings
[x] Any dependent changes have been merged and published

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[SamYuan1990]

+1
I am attempting strands agent with official strands agents example and use deep seek as LLM service.
I found some formatting issue, so, if by this PR, we can have official support with DeepSeek instead of OpenAI... that's great.

strands.types.exceptions.EventLoopException: Failed to deserialize the JSON body into the target type: messages[2]: invalid type: sequence, expected a string at line 1 column 755

[mo-silent]

Hi @veeragoni ,
I see the CICD checks are failing. Would you like help fixing these issues?
I'm also interested in adding DeepSeek support and could collaborate with you.

[pgrayy]

Deepseek provides an OpenAI compatible endpoint (docs) and so can already be accessed using Strand's OpenAIModel provider. You just need to pass in the base_url into client_args (docs).

[mo-silent]

Deepseek provides an OpenAI compatible endpoint (docs) and so can already be accessed using Strand's OpenAIModel provider. You just need to pass in the base_url into client_args (docs).

@pgrayy
I think you haven't tried to call Deepseek using the OpenAIModel provider. Deepseek does offer endpoints compatible with OpenAI. But in the OpenAIModel definition of stands-agent, it encapsulates request, which makes it incompatible with deepseek. For specific differences, you can refer to my screenshot:

Processing of agent OpenAIModel

Openai chat create api reference

Deepseek chat create api reference

Solution

1. Add Deepseek module provider
2. Create a Custom Model Provider, inherit from OpenAIModel Provider and add subclasses to handle the content method

However, I have found that many people have the need to call deepseek, so I think it would be better to add a deepseek provider

[afarntrog]

Hi @veeragoni !
Thank you so much for this PR and your contribution! We really appreciate the work you've put into this.
While we'd love to include every great model provider, we've reached a point where we need to be selective about adding new providers directly to the core SDK to keep it maintainable and focused.

That said, we absolutely want to support your integration! We'd recommend you publish the model provider implementation as a standalone package on PyPI. This way, customers can easily install it and use it with our SDK. Here's a great example of how this works: https://pypi.org/project/strands-nvidia-nim/

We'd be happy to feature your implementation on our documentation page as a supported model provider, giving you visibility to our community.

We're excited to see a DeepSeek model provider driven by the community!