AI & Agent Dev Bug Sandbox Back to Radar

ChatDeepSeek With Use_responses_api=True Raises KeyError('Messages')

ChatDeepSeek._get_request_payload() unconditionally accesses payload['messages'], but Responses API payloads use 'input'. Additionally, streaming does not delegate to Responses streaming methods. This causes all invoke/ainvoke/stream/astream calls to fail before any HTTP request.

highConfidence 92%Langchain Affected V1.1.0

Origin Analysis

The DeepSeek override of _get_request_payload assumes a Chat Completions payload containing 'messages'. When use_responses_api (or output_version='responses/v1') is enabled, BaseChatOpenAI._get_request_payload returns a Responses payload with 'input', so the key lookup raises KeyError. The class also does not override _stream/_astream to route to _stream_responses/_astream_responses.

Install langchain-deepseek 1.1.0 with langchain-openai 1.5.2a1. Run: from langchain_deepseek import ChatDeepSeek model = ChatDeepSeek(model='deepseek-v4-flash', api_key='test', base_url='https://example.invalid/v1', use_responses_api=True) model.invoke('Reply exactly OK') Observe KeyError: 'messages' before any HTTP request. The same occurs with stream(), ainvoke(), and astream().

Fixing Code Block

0x7F 45 4C 46 02 01 01 000x90 12 AB 04 C1 7E 33 08
0xA4 1C 90 3E 77 0B D2 580x11 F6 6A 2C 80 19 DE 44
0x00 00 00 00 03 00 3E 000x01 00 B7 4D 55 90 0C 61
0x6C 69 62 63 2E 73 6F 2E0x36 00 8E 17 42 AA 05 F3
0xDE C0 DE 01 8B 45 FC 890xC7 5D 20 6F 91 03 E4 78
0x13 37 00 4F 9A 2D E1 700x08 B4 5C 22 71 D0 4A 16
0x48 8D 3D 22 01 00 00 E80x91 FF 0A 63 2E 57 B8 04
0x5A 91 C3 0E 44 70 1B 880x02 6F 39 C4 15 80 ED 27
0xFF 15 40 20 60 00 48 890xC3 90 7A 01 4C 28 B1 56
0x0F 1F 44 00 00 55 48 890xE5 C3 18 6D 92 0B 33 A7

Edge Case Audit

This fix assumes that ChatDeepSeek's only normalization is removing the 'name' field. If your version includes additional message cleanups, preserve them inside the 'messages' guard. The streaming overrides must exactly match the base class method signatures or upgrade of langchain-openai may break. After applying, Responses API requests will reach DeepSeek's endpoint; if DeepSeek does not truly support the Responses API, you will get a server-side 4xx/5xx instead of a client-side KeyError, which is expected but still requires model support. Rollback by removing the added methods and reverting _get_request_payload to its original unconditional loop.