is it possible to provide a detail step by step guide how to use together with deepseek? thanks

[tonywang78]

I tried to setup it to use deepseek but it seems it can not invoke the tool correctly. the right sidebar display nth.
here is the error messages:

2025-04-27 21:14:19,784 - ERROR - Error processing stream: cannot access local variable 'complete_native_tool_calls' where it is not associated with a value
Traceback (most recent call last):
File "C:\Users\tonyw\Documents\GitHub\suna\backend\agentpress\response_processor.py", line 406, in process_streaming_response
if config.native_tool_calling and complete_native_tool_calls:
^^^^^^^^^^^^^^^^^^^^^^^^^^
UnboundLocalError: cannot access local variable 'complete_native_tool_calls' where it is not associated with a value
2025-04-27 21:14:20,638 - INFO - Successfully added message to thread 73b689db-6c25-4051-b83f-49f40736e05c
2025-04-27 21:14:20,648 - INFO - Stream generator main loop cancelled for b1ced9b7-0833-4227-bb1e-f669393ee8ab
2025-04-27 21:14:20,649 - ERROR - Error in listener for b1ced9b7-0833-4227-bb1e-f669393ee8ab: Connection closed by server.

[sahariarpku]

how did you connect deepseek?

[tonywang78]

Thanks for the reply, here is what i have done:

1. agent/api.py add deepseek mapping.

"deepseek-chat": "openai/deepseek-chat",
"deepseek-reasoning": "openai/deepseek-reasoning",

2. agent/run.py switch the tool calling:

                xml_tool_calling=False,
                native_tool_calling=True,

3. chat-input.tsx add deepseek option and update the default state to deepseek-chat (it seems the current version didn't provide the model switch button yet or i missed it somewhere):
   { id: "deepseek-chat", label: "Deepseek v3"},
   { id: "deepseek-reasoning", label: "deepseek r1"}

4. update .env file for the openai_api_base, openai_api_key and model_to_use

[sahariarpku]

I've added the initialization of complete_native_tool_calls as an empty list at the beginning of the method. This should fix the error you're seeing.
The error was occurring because the code was trying to access complete_native_tool_calls in the condition:

// ... existing code ...
accumulated_content = ""
tool_calls_buffer = {}
current_xml_content = ""
xml_chunks_buffer = []
pending_tool_executions = []
yielded_tool_indices = set() # Stores indices of tools whose status has been yielded
tool_index = 0
xml_tool_call_count = 0
finish_reason = None
last_assistant_message_object = None # Store the final saved assistant message object
tool_result_message_objects = {} # tool_index -> full saved message object
has_printed_thinking_prefix = False # Flag for printing thinking prefix only once
complete_native_tool_calls = [] # Store complete native tool calls for processing
// ... existing code ...

[sahariarpku]

also, pip3 install uvicorn fastapi

[tonywang78]

thank you @sahariarpku, it's working now.

[sahariarpku]

@tonywang78 did get to run whole thing with deepseek? Mine one is still having an issue! Would you mind share ur version through wechat please? My wechat is sahariar99

[adrianhsm]

I could use deepseek in suna now, what I do is:

1. set openai_api_key and MODEL_TO_USE in .env
   OPENAI_API_KEY="xxxxxxxx"
   MODEL_TO_USE="deepseek-chat"
2. add "deepseek-chat": "openai/deepseek-chat" to MODEL_NAME_ALIASES in api.py
3. set api_base="https://api.deepseek.com" for params = prepare_params(.....) in make_llm_api_call in llm.py for temporarily use

[markokraemer]

Is available out of the box now with the model switcher.

[byliu]

Is available out of the box now with the model switcher.

how? @markokraemer