diff --git a/flutter/lib/models/terminal_model.dart b/flutter/lib/models/terminal_model.dart
index a74241ccb..77198a4e4 100644
--- a/flutter/lib/models/terminal_model.dart
+++ b/flutter/lib/models/terminal_model.dart
@@ -31,6 +31,8 @@ class TerminalModel with ChangeNotifier {
   static const int _kMaxOutputBufferChars = 8 * 1024;
   // View ready state: true when terminal has valid dimensions, safe to write
   bool _terminalViewReady = false;
+  // Buffer for incomplete UTF-8 trailing bytes across data chunks.
+  final _utf8Remainder = <int>[];
 
   bool get isPeerWindows => parent.ffiModel.pi.platform == kPeerPlatformWindows;
 
@@ -333,31 +335,80 @@ class TerminalModel with ChangeNotifier {
 
     if (data != null) {
       try {
-        String text = '';
+        List<int> bytes;
         if (data is String) {
           // Try to decode as base64 first
           try {
-            final bytes = base64Decode(data);
-            text = utf8.decode(bytes, allowMalformed: true);
+            bytes = base64Decode(data);
           } catch (e) {
             // If base64 decode fails, treat as plain text
-            text = data;
+            _writeToTerminal(data);
+            return;
           }
         } else if (data is List) {
-          // Handle if data comes as byte array
-          text = utf8.decode(List<int>.from(data), allowMalformed: true);
+          bytes = List<int>.from(data);
         } else {
           debugPrint('[TerminalModel] Unknown data type: ${data.runtimeType}');
           return;
         }
 
-        _writeToTerminal(text);
+        // Prepend any leftover bytes from previous chunk.
+        if (_utf8Remainder.isNotEmpty) {
+          bytes = [..._utf8Remainder, ...bytes];
+          _utf8Remainder.clear();
+        }
+
+        // Find the split point so we only decode complete UTF-8 sequences.
+        final split = _findUtf8SplitPoint(bytes);
+        if (split < bytes.length) {
+          _utf8Remainder.addAll(bytes.sublist(split));
+        }
+
+        if (split > 0) {
+          final text = utf8.decode(bytes.sublist(0, split));
+          _writeToTerminal(text);
+        }
       } catch (e) {
         debugPrint('[TerminalModel] Failed to process terminal data: $e');
       }
     }
   }
 
+  /// Find the largest prefix length that ends on a complete UTF-8 character
+  /// boundary. Returns [bytes.length] if no trailing incomplete sequence.
+  static int _findUtf8SplitPoint(List<int> bytes) {
+    if (bytes.isEmpty) return 0;
+    // Inspect at most the last 3 bytes (max incomplete tail of 4-byte seq).
+    final start = bytes.length >= 3 ? bytes.length - 3 : 0;
+    for (int i = bytes.length - 1; i >= start; i--) {
+      final b = bytes[i];
+      if (b & 0x80 == 0) {
+        // ASCII – everything is complete.
+        return bytes.length;
+      }
+      if (b & 0xC0 == 0x80) {
+        // Continuation byte – keep scanning backwards.
+        continue;
+      }
+      // Leading byte – determine expected sequence length.
+      int seqLen;
+      if (b & 0xE0 == 0xC0) {
+        seqLen = 2;
+      } else if (b & 0xF0 == 0xE0) {
+        seqLen = 3;
+      } else if (b & 0xF8 == 0xF0) {
+        seqLen = 4;
+      } else {
+        // Invalid leading byte – treat as complete.
+        return bytes.length;
+      }
+      final available = bytes.length - i;
+      return available >= seqLen ? bytes.length : i;
+    }
+    // All inspected bytes are continuation bytes – treat as complete.
+    return bytes.length;
+  }
+
   /// Write text to terminal, buffering if the view is not yet ready.
   /// All terminal output should go through this method to avoid NaN errors
   /// from writing before the terminal view has valid layout dimensions.
@@ -427,6 +478,7 @@ class TerminalModel with ChangeNotifier {
     _inputBuffer.clear();
     _pendingOutputChunks.clear();
     _pendingOutputSize = 0;
+    _utf8Remainder.clear();
     // Terminal cleanup is handled server-side when service closes
     super.dispose();
   }
diff --git a/src/server/terminal_service.rs b/src/server/terminal_service.rs
index fb6b4fd29..74fddec7c 100644
--- a/src/server/terminal_service.rs
+++ b/src/server/terminal_service.rs
@@ -531,6 +531,54 @@ impl OutputBuffer {
     }
 }
 
+/// Find the largest prefix of `buf` that is valid UTF-8, i.e. does not end
+/// with an incomplete multi-byte sequence.  Returns `buf.len()` when the
+/// entire buffer is complete.  The caller should send `buf[..split]` and
+/// keep `buf[split..]` as a remainder for the next read.
+///
+/// We only need to inspect the last 1-3 bytes because the longest UTF-8
+/// sequence is 4 bytes.
+fn find_utf8_split_point(buf: &[u8]) -> usize {
+    if buf.is_empty() {
+        return 0;
+    }
+    // Walk backwards up to 3 bytes (max incomplete tail of a 4-byte sequence).
+    let start = if buf.len() >= 3 { buf.len() - 3 } else { 0 };
+    for i in (start..buf.len()).rev() {
+        let b = buf[i];
+        if b & 0x80 == 0 {
+            // ASCII – everything up to and including this byte is complete.
+            return buf.len();
+        }
+        // Continuation byte (10xxxxxx) – keep scanning back.
+        if b & 0xC0 == 0x80 {
+            continue;
+        }
+        // Leading byte – determine expected sequence length.
+        let seq_len = if b & 0xE0 == 0xC0 {
+            2
+        } else if b & 0xF0 == 0xE0 {
+            3
+        } else if b & 0xF8 == 0xF0 {
+            4
+        } else {
+            // Invalid leading byte – treat as complete to avoid stalling.
+            return buf.len();
+        };
+        let available = buf.len() - i;
+        if available >= seq_len {
+            // Sequence is complete.
+            return buf.len();
+        } else {
+            // Incomplete sequence starts at `i`.
+            return i;
+        }
+    }
+    // All inspected bytes are continuation bytes without a leading byte –
+    // treat as complete (probably binary / non-UTF-8 data).
+    buf.len()
+}
+
 /// Try to send data through the output channel with rate-limited drop logging.
 /// Returns `true` if the caller should break out of the read loop (channel disconnected).
 fn try_send_output(
@@ -1086,22 +1134,52 @@ impl TerminalServiceProxy {
         let reader_thread = thread::spawn(move || {
             let mut reader = reader;
             let mut buf = vec![0u8; 4096];
+            let mut utf8_remainder: Vec<u8> = Vec::new();
             let mut drop_count: u64 = 0;
             // Initialize to > 5s ago so the first drop triggers a warning immediately.
             let mut last_drop_warn = Instant::now() - Duration::from_secs(6);
             loop {
                 match reader.read(&mut buf) {
                     Ok(0) => {
-                        // EOF
-                        // This branch can be reached when the child process exits on macOS.
-                        // But not on Linux and Windows in my tests.
+                        // EOF – flush any leftover remainder bytes.
+                        if !utf8_remainder.is_empty() {
+                            let _ = try_send_output(
+                                &output_tx,
+                                utf8_remainder.split_off(0),
+                                terminal_id,
+                                "",
+                                &mut drop_count,
+                                &mut last_drop_warn,
+                            );
+                        }
                         break;
                     }
                     Ok(n) => {
                         if exiting.load(Ordering::SeqCst) {
                             break;
                         }
-                        let data = buf[..n].to_vec();
+                        // Prepend any leftover bytes from the previous read.
+                        let combined = if utf8_remainder.is_empty() {
+                            buf[..n].to_vec()
+                        } else {
+                            let mut v = std::mem::take(&mut utf8_remainder);
+                            v.extend_from_slice(&buf[..n]);
+                            v
+                        };
+                        // Split at a UTF-8 character boundary so we never send
+                        // an incomplete multi-byte sequence.
+                        let split = find_utf8_split_point(&combined);
+                        if split < combined.len() {
+                            utf8_remainder = combined[split..].to_vec();
+                        }
+                        let data = if split == combined.len() {
+                            combined
+                        } else {
+                            combined[..split].to_vec()
+                        };
+                        if data.is_empty() {
+                            continue;
+                        }
                         // Use try_send to avoid blocking the reader thread when channel is full.
                         // During disconnect, the run loop (sp.ok()) stops and read_outputs() is
                         // no longer called, so the channel won't be drained. Blocking send would
@@ -1308,13 +1386,24 @@ impl TerminalServiceProxy {
         let terminal_id = open.terminal_id;
         let reader_thread = thread::spawn(move || {
             let mut buf = vec![0u8; 4096];
+            let mut utf8_remainder: Vec<u8> = Vec::new();
             let mut drop_count: u64 = 0;
             // Initialize to > 5s ago so the first drop triggers a warning immediately.
             let mut last_drop_warn = Instant::now() - Duration::from_secs(6);
             loop {
                 match output_pipe.read(&mut buf) {
                     Ok(0) => {
-                        // EOF - helper process exited
+                        // EOF – flush any leftover remainder bytes.
+                        if !utf8_remainder.is_empty() {
+                            let _ = try_send_output(
+                                &output_tx,
+                                utf8_remainder.split_off(0),
+                                terminal_id,
+                                " (helper)",
+                                &mut drop_count,
+                                &mut last_drop_warn,
+                            );
+                        }
                         log::debug!("Terminal {} helper output EOF", terminal_id);
                         break;
                     }
@@ -1322,7 +1411,26 @@ impl TerminalServiceProxy {
                         if exiting.load(Ordering::SeqCst) {
                             break;
                         }
-                        let data = buf[..n].to_vec();
+                        // Prepend any leftover bytes from the previous read.
+                        let combined = if utf8_remainder.is_empty() {
+                            buf[..n].to_vec()
+                        } else {
+                            let mut v = std::mem::take(&mut utf8_remainder);
+                            v.extend_from_slice(&buf[..n]);
+                            v
+                        };
+                        let split = find_utf8_split_point(&combined);
+                        if split < combined.len() {
+                            utf8_remainder = combined[split..].to_vec();
+                        }
+                        let data = if split == combined.len() {
+                            combined
+                        } else {
+                            combined[..split].to_vec()
+                        };
+                        if data.is_empty() {
+                            continue;
+                        }
                         // Use try_send to avoid blocking the reader thread (same as direct PTY mode)
                         if try_send_output(
                             &output_tx,
@@ -1845,3 +1953,123 @@ impl TerminalServiceProxy {
         }
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_utf8_split_empty() {
+        assert_eq!(find_utf8_split_point(&[]), 0);
+    }
+
+    #[test]
+    fn test_utf8_split_ascii_only() {
+        assert_eq!(find_utf8_split_point(b"hello"), 5);
+    }
+
+    #[test]
+    fn test_utf8_split_complete_2byte() {
+        // 'é' = 0xC3 0xA9
+        let data = "café".as_bytes();
+        assert_eq!(find_utf8_split_point(data), data.len());
+    }
+
+    #[test]
+    fn test_utf8_split_complete_3byte() {
+        // '中' = 0xE4 0xB8 0xAD
+        let data = "中文".as_bytes();
+        assert_eq!(find_utf8_split_point(data), data.len());
+    }
+
+    #[test]
+    fn test_utf8_split_complete_4byte() {
+        // '😀' = 0xF0 0x9F 0x98 0x80
+        let data = "😀".as_bytes();
+        assert_eq!(find_utf8_split_point(data), data.len());
+    }
+
+    #[test]
+    fn test_utf8_split_incomplete_2byte() {
+        // Leading byte of 2-byte seq without continuation
+        let data = vec![b'a', 0xC3];
+        assert_eq!(find_utf8_split_point(&data), 1);
+    }
+
+    #[test]
+    fn test_utf8_split_incomplete_3byte_1of3() {
+        // Leading byte of 3-byte seq, missing 2 continuation bytes
+        let data = vec![b'a', 0xE4];
+        assert_eq!(find_utf8_split_point(&data), 1);
+    }
+
+    #[test]
+    fn test_utf8_split_incomplete_3byte_2of3() {
+        // Leading byte of 3-byte seq + 1 continuation, missing 1
+        let data = vec![b'a', 0xE4, 0xB8];
+        assert_eq!(find_utf8_split_point(&data), 1);
+    }
+
+    #[test]
+    fn test_utf8_split_incomplete_4byte_1of4() {
+        let data = vec![b'a', 0xF0];
+        assert_eq!(find_utf8_split_point(&data), 1);
+    }
+
+    #[test]
+    fn test_utf8_split_incomplete_4byte_2of4() {
+        let data = vec![b'a', 0xF0, 0x9F];
+        assert_eq!(find_utf8_split_point(&data), 1);
+    }
+
+    #[test]
+    fn test_utf8_split_incomplete_4byte_3of4() {
+        let data = vec![b'a', 0xF0, 0x9F, 0x98];
+        assert_eq!(find_utf8_split_point(&data), 1);
+    }
+
+    #[test]
+    fn test_utf8_split_mixed_complete() {
+        // ASCII + 3-byte (中) + ASCII
+        let data = "a中b".as_bytes();
+        assert_eq!(find_utf8_split_point(data), data.len());
+    }
+
+    #[test]
+    fn test_utf8_split_trailing_incomplete_after_valid() {
+        // "中" (complete) followed by incomplete start of another 3-byte char
+        let mut data = "中".as_bytes().to_vec();
+        data.push(0xE4); // leading byte of next 3-byte char
+        assert_eq!(find_utf8_split_point(&data), 3); // split before incomplete
+    }
+
+    #[test]
+    fn test_utf8_split_only_continuation_bytes() {
+        // Orphan continuation bytes (no leading byte) – treated as complete
+        let data = vec![0x80, 0x81, 0x82];
+        assert_eq!(find_utf8_split_point(&data), data.len());
+    }
+
+    #[test]
+    fn test_utf8_split_simulated_chunked_read() {
+        // Simulate reading "你好世界" in awkward chunk boundaries
+        let full = "你好世界".as_bytes().to_vec(); // 12 bytes (3 * 4 chars)
+        assert_eq!(full.len(), 12);
+
+        let mut decoded = Vec::new();
+        let mut remainder: Vec<u8> = Vec::new();
+
+        // Read in chunks of 5 bytes (will split mid-character)
+        for chunk in full.chunks(5) {
+            let mut combined = remainder.clone();
+            combined.extend_from_slice(chunk);
+            let split = find_utf8_split_point(&combined);
+            decoded.extend_from_slice(&combined[..split]);
+            remainder = combined[split..].to_vec();
+        }
+        // Flush remainder
+        decoded.extend_from_slice(&remainder);
+
+        assert_eq!(String::from_utf8(decoded).unwrap(), "你好世界");
+    }
+}