feat: parse and expose DeepSeek's reasoning content (#10121)

Co-authored-by: Henry Fontanier <henry@dust.tt>

Author: fontanierh

core/src/providers/anthropic.rs

@@ -557,6 +557,7 @@ impl TryFrom<ChatResponse> for AssistantChatMessage {
             role: ChatMessageRole::Assistant,
             name: None,
             content: text_content,
+            reasoning_content: None,
             function_call,
             function_calls,
         })

core/src/providers/chat_messages.rs

@@ -76,6 +76,8 @@ pub struct AssistantChatMessage {
     #[serde(skip_serializing_if = "Option::is_none")]
     pub content: Option<String>,
     #[serde(skip_serializing_if = "Option::is_none")]
+    pub reasoning_content: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
     pub function_call: Option<ChatFunctionCall>,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub function_calls: Option<Vec<ChatFunctionCall>>,

core/src/providers/mistral.rs

@@ -237,6 +237,7 @@ impl TryFrom<&MistralChatMessage> for AssistantChatMessage {
 
         Ok(AssistantChatMessage {
             content,
+            reasoning_content: None,
             role,
             name: None,
             function_call,

core/src/providers/openai_compatible_helpers.rs

@@ -252,6 +252,8 @@ pub struct OpenAIChatMessage {
     #[serde(skip_serializing_if = "Option::is_none")]
     pub content: Option<OpenAIChatMessageContent>,
     #[serde(skip_serializing_if = "Option::is_none")]
+    pub reasoning_content: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
     pub name: Option<String>,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub tool_calls: Option<Vec<OpenAIToolCall>>,
@@ -264,6 +266,8 @@ pub struct OpenAICompletionChatMessage {
     #[serde(skip_serializing_if = "Option::is_none")]
     pub content: Option<String>,
     #[serde(skip_serializing_if = "Option::is_none")]
+    pub reasoning_content: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
     pub name: Option<String>,
     pub role: OpenAIChatMessageRole,
     #[serde(skip_serializing_if = "Option::is_none")]
@@ -331,6 +335,10 @@ impl TryFrom<&OpenAICompletionChatMessage> for AssistantChatMessage {
             Some(c) => Some(c.clone()),
             None => None,
         };
+        let reasoning_content = match cm.reasoning_content.as_ref() {
+            Some(c) => Some(c.clone()),
+            None => None,
+        };
 
         let function_calls = if let Some(tool_calls) = cm.tool_calls.as_ref() {
             let cfc = tool_calls
@@ -360,6 +368,7 @@ impl TryFrom<&OpenAICompletionChatMessage> for AssistantChatMessage {
 
         Ok(AssistantChatMessage {
             content,
+            reasoning_content,
             role,
             name,
             function_call,
@@ -429,6 +438,7 @@ impl TryFrom<&ChatMessage> for OpenAIChatMessage {
                     Some(c) => Some(OpenAIChatMessageContent::try_from(c)?),
                     None => None,
                 },
+                reasoning_content: None,
                 name: assistant_msg.name.clone(),
                 role: OpenAIChatMessageRole::from(&assistant_msg.role),
                 tool_calls: match assistant_msg.function_calls.as_ref() {
@@ -443,20 +453,23 @@ impl TryFrom<&ChatMessage> for OpenAIChatMessage {
             }),
             ChatMessage::Function(function_msg) => Ok(OpenAIChatMessage {
                 content: Some(OpenAIChatMessageContent::try_from(&function_msg.content)?),
+                reasoning_content: None,
                 name: None,
                 role: OpenAIChatMessageRole::Tool,
                 tool_calls: None,
                 tool_call_id: Some(function_msg.function_call_id.clone()),
             }),
             ChatMessage::System(system_msg) => Ok(OpenAIChatMessage {
                 content: Some(OpenAIChatMessageContent::try_from(&system_msg.content)?),
+                reasoning_content: None,
                 name: None,
                 role: OpenAIChatMessageRole::from(&system_msg.role),
                 tool_calls: None,
                 tool_call_id: None,
             }),
             ChatMessage::User(user_msg) => Ok(OpenAIChatMessage {
                 content: Some(OpenAIChatMessageContent::try_from(&user_msg.content)?),
+                reasoning_content: None,
                 name: user_msg.name.clone(),
                 role: OpenAIChatMessageRole::from(&user_msg.role),
                 tool_calls: None,
@@ -806,6 +819,7 @@ fn to_openai_messages(
                         // Case 3: there's more than one content, the content isn't text or we don't want to squash them => keep structured format
                         (_, _, _) => Some(OpenAIChatMessageContent::Structured(contents)),
                     },
+                    reasoning_content: None,
                 }
             }
         })
@@ -829,6 +843,7 @@ fn to_openai_messages(
                         .collect()
                 }),
                 content: m.content,
+                reasoning_content: None,
             }
         })
         // Remove system messages if requested.
@@ -1215,6 +1230,7 @@ async fn streamed_chat_completion(
                 .map(|c| OpenAIChatChoice {
                     message: OpenAICompletionChatMessage {
                         content: Some("".to_string()),
+                        reasoning_content: None,
                         name: None,
                         role: OpenAIChatMessageRole::System,
                         tool_calls: None,
@@ -1272,6 +1288,14 @@ async fn streamed_chat_completion(
                     },
                 };
 
+                match a.choices[j].delta.get("reasoning_content") {
+                    None => (),
+                    Some(reasoning_content) => {
+                        c.choices[j].message.reasoning_content =
+                            Some(reasoning_content.as_str().unwrap_or("").to_string());
+                    }
+                };
+
                 if let Some(tool_calls) = a.choices[j]
                     .delta
                     .get("tool_calls")
@@ -1346,6 +1370,10 @@ async fn streamed_chat_completion(
             None => None,
             Some(c) => Some(c.trim().to_string()),
         };
+        m.message.reasoning_content = match m.message.reasoning_content.as_ref() {
+            None => None,
+            Some(c) => Some(c.trim().to_string()),
+        };
     }
 
     Ok((completion, request_id))
@@ -1497,6 +1525,10 @@ async fn chat_completion(
             None => None,
             Some(c) => Some(c.trim().to_string()),
         };
+        m.message.reasoning_content = match m.message.reasoning_content.as_ref() {
+            None => None,
+            Some(c) => Some(c.trim().to_string()),
+        };
     }
 
     Ok((completion, request_id))