diff --git a/src/kimi_cli/ui/shell/prompt.py b/src/kimi_cli/ui/shell/prompt.py
index b820fe8a00..a3ba813ed8 100644
--- a/src/kimi_cli/ui/shell/prompt.py
+++ b/src/kimi_cli/ui/shell/prompt.py
@@ -647,6 +647,7 @@ def __init__(
         self._cache_time: float = 0.0
         self._cached_paths: list[str] = []
         self._cache_scope: str | None = None
+        self._cache_query: str | None = None
         self._top_cache_time: float = 0.0
         self._top_cached_paths: list[str] = []
         self._fragment_hint: str | None = None
@@ -712,6 +713,8 @@ def _get_deep_paths(self) -> list[str]:
         cache_valid = (
             now - self._cache_time <= self._refresh_interval and self._cache_scope == scope
         )
+        if cache_valid and self._is_git is False and scope is None:
+            cache_valid = self._cache_query == fragment
 
         # Invalidate on .git/index mtime change (like Claude Code).
         if cache_valid and self._is_git:
@@ -730,10 +733,16 @@ def _get_deep_paths(self) -> list[str]:
             paths = list_files_git(self._root, scope)
             self._git_index_mtime = git_index_mtime(self._root)
         if paths is None:
-            paths = list_files_walk(self._root, scope, limit=self._limit)
+            paths = list_files_walk(
+                self._root,
+                scope,
+                limit=self._limit,
+                query=fragment if scope is None else None,
+            )
 
         self._cached_paths = paths
         self._cache_scope = scope
+        self._cache_query = fragment if self._is_git is False and scope is None else None
         self._cache_time = now
         return self._cached_paths
 
diff --git a/src/kimi_cli/utils/file_filter.py b/src/kimi_cli/utils/file_filter.py
index 64ac127724..be32de4f97 100644
--- a/src/kimi_cli/utils/file_filter.py
+++ b/src/kimi_cli/utils/file_filter.py
@@ -3,6 +3,7 @@
 import os
 import re
 import subprocess
+from collections import deque
 from pathlib import Path
 
 _IGNORED_NAMES: frozenset[str] = frozenset(
@@ -293,10 +294,16 @@ def list_files_walk(
     scope: str | None = None,
     *,
     limit: int = 1000,
+    query: str | None = None,
+    scan_limit: int | None = None,
 ) -> list[str]:
     """List workspace paths via ``os.walk`` (fallback for non-git repos).
 
     When *scope* is given, the walk starts from that subdirectory.
+    When *query* is given, only fuzzy subsequence matches are returned.  The
+    scan remains bounded independently from the result limit so selective
+    queries can find entries beyond the first result page without turning a
+    completion request into an unbounded workspace walk.
     """
     resolved_root = root.resolve()
     walk_root = (root / scope).resolve() if scope else resolved_root
@@ -308,6 +315,15 @@ def list_files_walk(
     except (OSError, ValueError):
         return []
 
+    if query:
+        return _list_files_walk_query(
+            resolved_root,
+            walk_root,
+            query=query,
+            limit=limit,
+            scan_limit=scan_limit if scan_limit is not None else limit * 10,
+        )
+
     paths: list[str] = []
     try:
         for current_root, dirs, files in os.walk(walk_root):
@@ -342,6 +358,52 @@ def list_files_walk(
     return paths
 
 
+def _list_files_walk_query(
+    resolved_root: Path,
+    walk_root: Path,
+    *,
+    query: str,
+    limit: int,
+    scan_limit: int,
+) -> list[str]:
+    """Return fuzzy matches from a bounded, lazy non-git workspace scan."""
+    normalized_query = query.casefold()
+    pending = deque([walk_root])
+    paths: list[str] = []
+    scanned = 0
+
+    def matches(path: str) -> bool:
+        chars = iter(path.casefold())
+        return all(any(char == candidate for candidate in chars) for char in normalized_query)
+
+    while pending and scanned < scan_limit and len(paths) < limit:
+        directory = pending.popleft()
+        try:
+            with os.scandir(directory) as entries:
+                for entry in entries:
+                    scanned += 1
+                    if scanned > scan_limit:
+                        break
+                    if is_ignored(entry.name):
+                        continue
+                    try:
+                        relative = Path(entry.path).relative_to(resolved_root).as_posix()
+                        is_dir = entry.is_dir(follow_symlinks=False)
+                    except (OSError, ValueError):
+                        continue
+                    if is_dir:
+                        pending.append(Path(entry.path))
+                        relative += "/"
+                    if matches(relative):
+                        paths.append(relative)
+                        if len(paths) >= limit:
+                            break
+        except OSError:
+            continue
+
+    return paths
+
+
 def list_directory_filtered(directory: Path) -> list[dict[str, str | int]]:
     """List immediate children of *directory*, filtering ignored entries.
 
diff --git a/tests/ui_and_conv/test_file_completer.py b/tests/ui_and_conv/test_file_completer.py
index 9df041ee44..d4bfe8fe81 100644
--- a/tests/ui_and_conv/test_file_completer.py
+++ b/tests/ui_and_conv/test_file_completer.py
@@ -120,6 +120,34 @@ def test_scoped_walk_finds_late_alphabetical_dirs(tmp_path: Path):
     assert "zzz_target/important.py" in texts
 
 
+def test_unscoped_query_finds_match_beyond_candidate_limit(tmp_path: Path):
+    """A selective query must search past the first candidate page.
+
+    Regression test for #1610: non-git workspaces used the 1000-result
+    candidate limit as a scan limit, so a matching file sorted after those
+    entries could never be completed.
+    """
+    for index in range(1100):
+        (tmp_path / f"aaa_{index:04d}.txt").write_text("")
+    target = tmp_path / "zzz_unique_needle.txt"
+    target.write_text("find me")
+
+    completer = LocalFileMentionCompleter(tmp_path, limit=1000)
+
+    texts = _completion_texts(completer, "@needle")
+
+    assert target.name in texts
+
+
+def test_unscoped_query_cache_is_keyed_by_fragment(tmp_path: Path):
+    (tmp_path / "unique_needle.txt").write_text("")
+    (tmp_path / "unique_other.txt").write_text("")
+    completer = LocalFileMentionCompleter(tmp_path)
+
+    assert "unique_needle.txt" in _completion_texts(completer, "@needle")
+    assert "unique_other.txt" in _completion_texts(completer, "@other")
+
+
 def test_basename_prefix_is_ranked_first(tmp_path: Path):
     """Prefer basename prefix matches over cross-segment fuzzy matches.
 
diff --git a/tests/utils/test_file_filter.py b/tests/utils/test_file_filter.py
index a5fd2fbc88..fecfef8f9f 100644
--- a/tests/utils/test_file_filter.py
+++ b/tests/utils/test_file_filter.py
@@ -15,6 +15,26 @@
 )
 
 
+def test_query_walk_uses_separate_bounded_scan_budget(tmp_path: Path) -> None:
+    for index in range(10):
+        (tmp_path / f"aaa_{index:02d}.txt").write_text("")
+    target = tmp_path / "zzz_needle.txt"
+    target.write_text("")
+
+    assert target.name not in list_files_walk(
+        tmp_path,
+        query="needle",
+        limit=1000,
+        scan_limit=10,
+    )
+    assert target.name in list_files_walk(
+        tmp_path,
+        query="needle",
+        limit=1000,
+        scan_limit=11,
+    )
+
+
 def _init_git(root: Path) -> None:
     for cmd in (
         ["git", "init"],