add Deepseek deepseek-chat model support

[lightionight]

current only add deepseek-chat(DeepSeek V3) support, if offical api update, will add deepseek-chat reasoner(DeepSeek R1) support.

[evalstate]

Hi Lightionight - this is definitely going to get merged - thank you so much for the contribution.

May I quickly check

1. Are you using https://www.deepseek.com/en as the API Provider? I want to add DeepSeek to the automated test suite so want to make sure I am testing against the same provider you are using.

I'm currently making some changes in this area to improve type safety and test coverage, so plan to merge this in to branch first, test and then publish.

Best, evalstate.

[lightionight]

Hi Lightionight - this is definitely going to get merged - thank you so much for the contribution.

May I quickly check

1. Are you using https://www.deepseek.com/en as the API Provider? I want to add DeepSeek to the automated test suite so want to make sure I am testing against the same provider you are using.

I'm currently making some changes in this area to improve type safety and test coverage, so plan to merge this in to branch first, test and then publish.

Best, evalstate.

Hello! evalstate

I’m glad to hear that. Yes, I am using this.

Currently, I have only added the DeepSeek-Chat model, as the Reasoner R1 model api
is still being adjusted by the official team.

Best lightionight

[evalstate]

Hi @lightionight -- quick update on this.

I've made quite a few updates in this area, and bought in a couple of your files and completed some testing. This is currently on "main" as I finish the 0.2.0 release notes.

Notes:

• I've created a new DeepSeekLLM here https://github.com/evalstate/fast-agent/blob/main/src/mcp_agent/llm/providers/augmented_llm_deepseek.py to reduce some of the duplication.
• I've added a couple of smoke tests for DeepSeek here: https://github.com/evalstate/fast-agent/blob/main/tests/e2e/smoke/test_e2e_smoke.py . Note that DeepSeek doesn't work with the beta OpenAI structured outputs features, so it falls back to the JSON parsing here :

  fast-agent/src/mcp_agent/llm/providers/augmented_llm_openai.py

  Lines 404 to 406 in 38b5ee5

  	if not "OpenAI" == self.provider:
  	return await super().structured(prompt, model, request_params)

• DeepSeek doesn't have vision capabilities. I'm about to add more precise Model configuration, so will tackle graceful handling there.

Let me know if you have the chance to take a look. We'll add the reasoning model once finalised?

[funnythingfunnylove]

Hi @lightionight -- quick update on this.

I've made quite a few updates in this area, and bought in a couple of your files and completed some testing. This is currently on "main" as I finish the 0.2.0 release notes.

Notes:

• I've created a new DeepSeekLLM here https://github.com/evalstate/fast-agent/blob/main/src/mcp_agent/llm/providers/augmented_llm_deepseek.py to reduce some of the duplication.
• I've added a couple of smoke tests for DeepSeek here: https://github.com/evalstate/fast-agent/blob/main/tests/e2e/smoke/test_e2e_smoke.py . Note that DeepSeek doesn't work with the beta OpenAI structured outputs features, so it falls back to the JSON parsing here :

  fast-agent/src/mcp_agent/llm/providers/augmented_llm_openai.py

  Lines 404 to 406 in 38b5ee5

  	if not "OpenAI" == self.provider:
  	return await super().structured(prompt, model, request_params)

• DeepSeek doesn't have vision capabilities. I'm about to add more precise Model configuration, so will tackle graceful handling there.

Let me know if you have the chance to take a look. We'll add the reasoning model once finalised?

Hi @evalstate nice work!
This almost done add deepseek api, Yes, i will contiune watch offical api changes, when it happend i will notice u, Thanks for your work
Best lightionight

[evalstate]

reopen/new PR after reasoning support tested.