diff --git a/src/api/completions.rs b/src/api/completions.rs
index 9208e66..0e09b47 100644
--- a/src/api/completions.rs
+++ b/src/api/completions.rs
@@ -212,36 +212,12 @@ async fn chat_completions_stream(
             if let Ok((status, data)) = state.backend.get_steps(&cascade_id).await {
                 if status == 200 {
                     if let Some(steps) = data["steps"].as_array() {
-                        let text = extract_response_text(steps);
-
-                        if !text.is_empty() && text != last_text {
-                            let delta = if text.len() > last_text.len() && text.starts_with(&*last_text) {
-                                &text[last_text.len()..]
-                            } else {
-                                &text
-                            };
-
-                            if !delta.is_empty() {
-                                yield Ok(Event::default().data(serde_json::to_string(&serde_json::json!({
-                                    "id": completion_id,
-                                    "object": "chat.completion.chunk",
-                                    "created": now_unix(),
-                                    "model": model_name,
-                                    "choices": [{
-                                        "index": 0,
-                                        "delta": {"content": delta},
-                                        "finish_reason": serde_json::Value::Null,
-                                    }],
-                                })).unwrap_or_default()));
-                                last_text = text.to_string();
-                            }
-                        }
-
-                        // Check for MITM-captured function calls (tool use)
+                        // Check for MITM-captured function calls FIRST (before text)
+                        // This prevents dummy placeholder text from leaking to client
                         let captured = state.mitm_store.take_any_function_calls().await;
                         if let Some(ref calls) = captured {
                             if !calls.is_empty() {
-                                // Emit tool_calls in OpenAI streaming format
+                                // Emit tool_calls in OpenAI streaming format — NO text
                                 let mut tool_calls = Vec::new();
                                 for (i, fc) in calls.iter().enumerate() {
                                     let call_id = format!(
@@ -288,6 +264,32 @@ async fn chat_completions_stream(
                             }
                         }
 
+                        // Normal text streaming (only when no function calls)
+                        let text = extract_response_text(steps);
+
+                        if !text.is_empty() && text != last_text {
+                            let delta = if text.len() > last_text.len() && text.starts_with(&*last_text) {
+                                &text[last_text.len()..]
+                            } else {
+                                &text
+                            };
+
+                            if !delta.is_empty() {
+                                yield Ok(Event::default().data(serde_json::to_string(&serde_json::json!({
+                                    "id": completion_id,
+                                    "object": "chat.completion.chunk",
+                                    "created": now_unix(),
+                                    "model": model_name,
+                                    "choices": [{
+                                        "index": 0,
+                                        "delta": {"content": delta},
+                                        "finish_reason": serde_json::Value::Null,
+                                    }],
+                                })).unwrap_or_default()));
+                                last_text = text.to_string();
+                            }
+                        }
+
                         // Done check: need DONE status AND non-empty text
                         if is_response_done(steps) && !last_text.is_empty() {
                             debug!("Completions stream done, text length={}", last_text.len());
diff --git a/src/mitm/proxy.rs b/src/mitm/proxy.rs
index 519c1ff..ae57e7b 100644
--- a/src/mitm/proxy.rs
+++ b/src/mitm/proxy.rs
@@ -850,14 +850,8 @@ async fn handle_http_over_tls(
         // Capture usage data
         if is_streaming_response {
             if streaming_acc.is_complete || streaming_acc.output_tokens > 0 {
-                // Save any captured function calls before consuming the accumulator
-                for fc in &streaming_acc.function_calls {
-                    store.record_function_call(cascade_hint.as_deref(), fc.clone()).await;
-                }
-                // Also save for history rewriting on tool result turns
-                if !streaming_acc.function_calls.is_empty() {
-                    store.set_last_function_calls(streaming_acc.function_calls.clone()).await;
-                }
+                // Function calls are stored immediately when detected (above),
+                // so no need to store them again here.
                 let usage = streaming_acc.into_usage();
                 store.record_usage(cascade_hint.as_deref(), usage).await;
             }