From 423d6378232eb82384bff9c35174df21122e1cee Mon Sep 17 00:00:00 2001
From: wjameswen888 <wjameswen888@users.noreply.github.com>
Date: Fri, 8 May 2026 12:53:26 +0900
Subject: [PATCH] fix(deepseek): send explicit thinking:disabled when reasoning
 is off
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

DeepSeek direct API requires an explicit thinking.type parameter.
When thinking is disabled (reasoning_effort=none or enabled=false),
the API defaults to thinking=enabled — generating reasoning_content
that gets stripped/not passed back, causing HTTP 400 on subsequent
turns.

This mirrors the existing Kimi extra_body.thinking pattern:
- Detect DeepSeek provider/model/base_url in run_agent.py
- Send {"thinking": {"type": "disabled"}} when appropriate
- Respect reasoning_config.effort=none as disabled

Fixes: #15700, #17212
Related: #15213
---
 agent/transports/chat_completions.py | 14 ++++++++++++++
 run_agent.py                         |  6 ++++++
 sessions.db                          |  0
 3 files changed, 20 insertions(+)
 create mode 100644 sessions.db

diff --git a/agent/transports/chat_completions.py b/agent/transports/chat_completions.py
index ca29b39ffe4e..1c21502679ca 100644
--- a/agent/transports/chat_completions.py
+++ b/agent/transports/chat_completions.py
@@ -258,6 +258,7 @@ def build_kwargs(
         anthropic_max_out = params.get("anthropic_max_output")
         is_nvidia_nim = params.get("is_nvidia_nim", False)
         is_kimi = params.get("is_kimi", False)
+        is_deepseek = params.get("is_deepseek", False)
         is_tokenhub = params.get("is_tokenhub", False)
         reasoning_config = params.get("reasoning_config")
 
@@ -333,6 +334,19 @@ def build_kwargs(
                 "type": "enabled" if _kimi_thinking_enabled else "disabled",
             }
 
+        # DeepSeek extra_body.thinking (same format as Kimi)
+        if is_deepseek:
+            _ds_thinking_enabled = True
+            if reasoning_config and isinstance(reasoning_config, dict):
+                if reasoning_config.get("enabled") is False:
+                    _ds_thinking_enabled = False
+                _ds_effort = (reasoning_config.get("effort") or "").strip().lower()
+                if _ds_effort == "none":
+                    _ds_thinking_enabled = False
+            extra_body["thinking"] = {
+                "type": "enabled" if _ds_thinking_enabled else "disabled",
+            }
+
         # Reasoning. LM Studio is handled above via top-level reasoning_effort,
         # so skip emitting extra_body.reasoning for it.
         if params.get("supports_reasoning", False) and not params.get("is_lmstudio", False):
diff --git a/run_agent.py b/run_agent.py
index 403dba4e7850..f00621ee14c2 100644
--- a/run_agent.py
+++ b/run_agent.py
@@ -8710,6 +8710,11 @@ def _build_api_kwargs(self, api_messages: list) -> dict:
             or base_url_host_matches(self.base_url, "moonshot.cn")
         )
         _is_tokenhub = base_url_host_matches(self._base_url_lower, "tokenhub.tencentmaas.com")
+        _is_deepseek = (
+            self.provider == "deepseek"
+            or "deepseek" in (self.model or "").lower()
+            or base_url_host_matches(self.base_url, "api.deepseek.com")
+        )
         _is_lmstudio = (self.provider or "").strip().lower() == "lmstudio"
 
         # Temperature: _fixed_temperature_for_model may return OMIT_TEMPERATURE
@@ -8819,6 +8824,7 @@ def _build_api_kwargs(self, api_messages: list) -> dict:
             is_github_models=_is_gh,
             is_nvidia_nim=_is_nvidia,
             is_kimi=_is_kimi,
+            is_deepseek=_is_deepseek,
             is_tokenhub=_is_tokenhub,
             is_lmstudio=_is_lmstudio,
             is_custom_provider=self.provider == "custom",
diff --git a/sessions.db b/sessions.db
new file mode 100644
index 000000000000..e69de29bb2d1