Add reasoning streaming

internal/llm/client_completions.go

@@ -177,18 +177,21 @@ func (c *Client) completions(
 
 // StreamEvent wraps a message or delta produced during streaming.
 type StreamEvent struct {
-	// Type identifies the event kind: delta, done, or message.
+	// Type identifies the event kind: delta, reasoning_delta, done, or message.
 	Type string
 	// Delta is a text fragment from the assistant response.
 	Delta string
+	// ReasoningDelta is a reasoning text fragment from the assistant response.
+	ReasoningDelta string
 	// Message is the complete chat completion message.
 	Message Message
 }
 
 const (
-	streamEventDelta   = "delta"
-	streamEventDone    = "done"
-	streamEventMessage = "message"
+	streamEventDelta          = "delta"
+	streamEventReasoningDelta = "reasoning_delta"
+	streamEventDone           = "done"
+	streamEventMessage        = "message"
 )
 
 // CompletionsStream sends messages to the LLM using the specified model and
@@ -403,6 +406,10 @@ func (c *Client) completionsStream(ctx context.Context, messages []Message, mode
 				acc.reasoning.WriteString(
 					delta.ReasoningContent,
 				)
+				events <- StreamEvent{
+					Type:           streamEventReasoningDelta,
+					ReasoningDelta: delta.ReasoningContent,
+				}
 			}
 			// Accumulate tool call deltas by index.
 			for _, tc := range delta.ToolCalls {