diff --git a/tests/tools/test_computer_use.py b/tests/tools/test_computer_use.py
index 7e7420db596..566e61aecee 100644
--- a/tests/tools/test_computer_use.py
+++ b/tests/tools/test_computer_use.py
@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import base64
 import json
 import os
 import sys
@@ -360,7 +361,9 @@ class TestCaptureResponse:
             def focus_app(self, app, raise_window=False): ...
 
         cu_tool.reset_backend_for_tests()
-        with patch.object(cu_tool, "_get_backend", return_value=FakeBackend()):
+        with patch.object(cu_tool, "_get_backend", return_value=FakeBackend()), \
+             patch.object(cu_tool, "_should_route_through_aux_vision",
+                          return_value=False):
             out = cu_tool.handle_computer_use({"action": "capture", "mode": "vision"})
 
         assert isinstance(out, dict)
@@ -398,7 +401,9 @@ class TestCaptureResponse:
             def focus_app(self, app, raise_window=False): ...
 
         cu_tool.reset_backend_for_tests()
-        with patch.object(cu_tool, "_get_backend", return_value=FakeBackend()):
+        with patch.object(cu_tool, "_get_backend", return_value=FakeBackend()), \
+             patch.object(cu_tool, "_should_route_through_aux_vision",
+                          return_value=False):
             out = cu_tool.handle_computer_use({"action": "capture", "mode": "som"})
         assert isinstance(out, dict)
         text_part = next(p for p in out["content"] if p.get("type") == "text")
@@ -436,6 +441,7 @@ class TestCaptureResponse:
 
         return FakeBackend()
 
+
     def test_capture_ax_caps_elements_at_default_for_dense_trees(self):
         """Regression for #22865: an Electron-style 600-element AX tree must
         not emit the entire array verbatim into the tool result.
@@ -582,7 +588,9 @@ class TestCaptureResponse:
             def focus_app(self, app, raise_window=False): ...
 
         cu_tool.reset_backend_for_tests()
-        with patch.object(cu_tool, "_get_backend", return_value=FakeBackend()):
+        with patch.object(cu_tool, "_get_backend", return_value=FakeBackend()), \
+             patch.object(cu_tool, "_should_route_through_aux_vision",
+                          return_value=False):
             out = cu_tool.handle_computer_use({"action": "capture", "mode": "som"})
 
         assert isinstance(out, dict) and out["_multimodal"] is True
@@ -594,6 +602,32 @@ class TestCaptureResponse:
         assert "truncated to" not in out["text_summary"]
 
 
+class TestCuaCaptureImageDimensions:
+    def test_png_dimensions_are_sniffed_from_image_bytes(self):
+        from tools.computer_use.cua_backend import _image_dimensions_from_bytes
+
+        raw_png = base64.b64decode(
+            "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42m"
+            "NkYAAAAAYAAjCB0C8AAAAASUVORK5CYII=",
+            validate=False,
+        )
+        assert _image_dimensions_from_bytes(raw_png) == (1, 1)
+
+    def test_jpeg_dimensions_are_sniffed_from_sof_segment(self):
+        from tools.computer_use.cua_backend import _image_dimensions_from_bytes
+
+        raw_jpeg = (
+            b"\xff\xd8" +
+            b"\xff\xe0\x00\x10" + (b"0" * 14)
+            + b"\xff\xc0\x00\x11\x08"
+            + b"\x01\x2c"  # height: 300
+            + b"\x01\x90"  # width: 400
+            + b"\x03\x01\x11\x00\x02\x11\x00\x03\x11\x00"
+            + b"\xff\xd9"
+        )
+        assert _image_dimensions_from_bytes(raw_jpeg) == (400, 300)
+
+
 # ---------------------------------------------------------------------------
 # Anthropic adapter: multimodal tool-result conversion
 # ---------------------------------------------------------------------------
diff --git a/tests/tools/test_computer_use_capture_routing.py b/tests/tools/test_computer_use_capture_routing.py
index 3eeb983ecea..b5a52356165 100644
--- a/tests/tools/test_computer_use_capture_routing.py
+++ b/tests/tools/test_computer_use_capture_routing.py
@@ -241,6 +241,39 @@ class TestCaptureResponseRoutedToAuxVision:
         assert observed_path["path"]
         assert not os.path.exists(observed_path["path"])
 
+    def test_aux_route_creates_missing_cache_dir(self, tmp_path):
+        from tools.computer_use import tool as cu_tool
+
+        cache_dir = tmp_path / "missing" / "cache_vision"
+        cap = _make_capture(mode="som")
+        observed_path = {}
+
+        def _fake_get(*_args, **_kw):
+            return cache_dir
+
+        def _fake_run_async(_coro):
+            return _stub_aux_analysis("description goes here")
+
+        def _fake_vat(image_path, _prompt):
+            observed_path["path"] = image_path
+            assert os.path.exists(image_path)
+            return "<coro>"
+
+        fake_vat = MagicMock(side_effect=_fake_vat)
+
+        with patch.object(cu_tool, "_should_route_through_aux_vision",
+                          return_value=True), \
+             patch("hermes_constants.get_hermes_dir", _fake_get), \
+             patch("model_tools._run_async", side_effect=_fake_run_async), \
+             patch("tools.vision_tools.vision_analyze_tool",
+                   new_callable=lambda: fake_vat):
+            resp = cu_tool._capture_response(cap)
+
+        assert isinstance(resp, str)
+        assert cache_dir.is_dir()
+        assert observed_path["path"]
+        assert not os.path.exists(observed_path["path"])
+
     def test_temp_file_cleaned_up_even_when_aux_call_raises(
         self, tmp_cache_dir,
     ):
diff --git a/tests/tools/test_computer_use_vision_routing.py b/tests/tools/test_computer_use_vision_routing.py
index b0ae4566994..3e3d4ee7df0 100644
--- a/tests/tools/test_computer_use_vision_routing.py
+++ b/tests/tools/test_computer_use_vision_routing.py
@@ -160,6 +160,42 @@ class TestRouteDecision:
                 "some-aggregator", "some-vision-model", {}
             ) is True
 
+    def test_user_declared_vision_support_keeps_custom_provider_native(self):
+        """Local/custom VLMs use config as their tool-result image escape hatch."""
+        from tools.computer_use import vision_routing
+
+        cfg = {
+            "model": {
+                "default": "Qwen3.6-35B-A3B-local-vlm",
+                "provider": "omlx",
+                "supports_vision": True,
+            }
+        }
+        with patch.object(vision_routing,
+                          "_provider_accepts_multimodal_tool_result",
+                          return_value=False):
+            assert vision_routing.should_route_capture_to_aux_vision(
+                "custom", "Qwen3.6-35B-A3B-local-vlm", cfg
+            ) is False
+
+    def test_user_declared_no_vision_routes_custom_provider_to_aux(self):
+        """An explicit false override should not fall through to native routing."""
+        from tools.computer_use import vision_routing
+
+        cfg = {
+            "model": {
+                "default": "local-text-model",
+                "provider": "omlx",
+                "supports_vision": False,
+            }
+        }
+        with patch.object(vision_routing,
+                          "_provider_accepts_multimodal_tool_result",
+                          return_value=True):
+            assert vision_routing.should_route_capture_to_aux_vision(
+                "custom", "local-text-model", cfg
+            ) is True
+
     def test_unknown_provider_capabilities_fail_closed(self):
         """When tool-result lookup returns None, route to aux (safe default)."""
         from tools.computer_use import vision_routing
diff --git a/tools/computer_use/cua_backend.py b/tools/computer_use/cua_backend.py
index 5ade2fdf85e..60c998c87d0 100644
--- a/tools/computer_use/cua_backend.py
+++ b/tools/computer_use/cua_backend.py
@@ -126,6 +126,45 @@ def _parse_elements_from_tree(markdown: str) -> List[UIElement]:
     return elements
 
 
+def _image_dimensions_from_bytes(raw: bytes) -> Tuple[int, int]:
+    """Best-effort PNG/JPEG dimension sniffing without extra dependencies."""
+    if raw.startswith(b"\x89PNG\r\n\x1a\n") and len(raw) >= 24:
+        width = int.from_bytes(raw[16:20], "big")
+        height = int.from_bytes(raw[20:24], "big")
+        if width > 0 and height > 0:
+            return width, height
+
+    if raw.startswith(b"\xff\xd8"):
+        i = 2
+        n = len(raw)
+        while i + 9 < n:
+            if raw[i] != 0xFF:
+                i += 1
+                continue
+            marker = raw[i + 1]
+            i += 2
+            if marker in {0xD8, 0xD9} or 0xD0 <= marker <= 0xD7:
+                continue
+            if i + 2 > n:
+                break
+            segment_len = int.from_bytes(raw[i:i + 2], "big")
+            if segment_len < 2 or i + segment_len > n:
+                break
+            if marker in {
+                0xC0, 0xC1, 0xC2, 0xC3, 0xC5, 0xC6, 0xC7,
+                0xC9, 0xCA, 0xCB, 0xCD, 0xCE, 0xCF,
+            }:
+                if segment_len >= 7:
+                    height = int.from_bytes(raw[i + 3:i + 5], "big")
+                    width = int.from_bytes(raw[i + 5:i + 7], "big")
+                    if width > 0 and height > 0:
+                        return width, height
+                break
+            i += segment_len
+
+    return 0, 0
+
+
 def _split_tree_text(full_text: str) -> Tuple[str, str]:
     """Split get_window_state text into (summary_line, tree_markdown)."""
     lines = full_text.split("\n", 1)
@@ -491,7 +530,12 @@ class CuaDriverBackend(ComputerUseBackend):
         png_bytes_len = 0
         if png_b64:
             try:
-                png_bytes_len = len(base64.b64decode(png_b64, validate=False))
+                raw = base64.b64decode(png_b64, validate=False)
+                png_bytes_len = len(raw)
+                detected_width, detected_height = _image_dimensions_from_bytes(raw)
+                if detected_width and detected_height:
+                    width = detected_width
+                    height = detected_height
             except Exception:
                 png_bytes_len = len(png_b64) * 3 // 4
 
diff --git a/tools/computer_use/tool.py b/tools/computer_use/tool.py
index abb14ebd878..7fa9c1de42f 100644
--- a/tools/computer_use/tool.py
+++ b/tools/computer_use/tool.py
@@ -615,6 +615,7 @@ def _route_capture_through_aux_vision(
         # MIME sniffing returns the right content-type.
         ext = ".jpg" if cap.png_b64[:8].startswith("/9j/") else ".png"
         cache_dir = get_hermes_dir("cache/vision", "temp_vision_images")
+        cache_dir.mkdir(parents=True, exist_ok=True)
         temp_image_path = cache_dir / f"computer_use_{_uuid.uuid4().hex}{ext}"
         temp_image_path.write_bytes(raw)
 
diff --git a/tools/computer_use/vision_routing.py b/tools/computer_use/vision_routing.py
index 3b4be1e15a6..5cc9e2fb61c 100644
--- a/tools/computer_use/vision_routing.py
+++ b/tools/computer_use/vision_routing.py
@@ -28,6 +28,10 @@ Behaviour (mirrors ``vision_analyze`` for consistency)
   ``provider``, ``model``, or ``base_url`` non-empty / not ``"auto"``),
   the screenshot is routed through the aux vision pipeline. Users who
   pay for a dedicated vision model usually want it used.
+* Otherwise, if the user explicitly declared the active model vision-capable
+  via ``model.supports_vision`` / provider model config, return ``False``.
+  This is the escape hatch for custom/local OpenAI-compatible VLM routes that
+  are absent from models.dev and provider allowlists.
 * Otherwise, if the active main model+provider can carry an image inside
   a tool-result message AND the model reports ``supports_vision=True``
   in models.dev metadata, return ``False`` (use the multimodal path).
@@ -76,10 +80,52 @@ def _explicit_aux_vision_override(cfg: Optional[Dict[str, Any]]) -> bool:
     return True
 
 
-def _lookup_supports_vision(provider: str, model: str) -> Optional[bool]:
-    """Return models.dev ``supports_vision`` for *(provider, model)* or None."""
+def _lookup_user_declared_supports_vision(
+    provider: str,
+    model: str,
+    cfg: Optional[Dict[str, Any]],
+) -> Optional[bool]:
+    """Return config-declared ``supports_vision`` for the active route."""
+    try:
+        from agent.image_routing import _supports_vision_override
+    except Exception as exc:  # pragma: no cover - defensive
+        logger.debug(
+            "computer_use vision_routing: config override lookup import failed: %s",
+            exc,
+        )
+        return None
+    try:
+        return _supports_vision_override(cfg, provider, model)
+    except Exception as exc:  # pragma: no cover - defensive
+        logger.debug(
+            "computer_use vision_routing: config override lookup failed: %s",
+            exc,
+        )
+        return None
+
+
+def _lookup_supports_vision(
+    provider: str,
+    model: str,
+    cfg: Optional[Dict[str, Any]] = None,
+) -> Optional[bool]:
+    """Return config/models.dev ``supports_vision`` for *(provider, model)*."""
     if not provider or not model:
         return None
+    try:
+        from agent.image_routing import _lookup_supports_vision as _lookup_image_supports
+    except Exception:
+        _lookup_image_supports = None
+    if _lookup_image_supports is not None:
+        try:
+            return _lookup_image_supports(provider, model, cfg)
+        except Exception as exc:  # pragma: no cover - defensive
+            logger.debug(
+                "computer_use vision_routing: image-routing caps lookup failed "
+                "for %s:%s — %s",
+                provider, model, exc,
+            )
+            return None
     try:
         from agent.models_dev import get_model_capabilities
         caps = get_model_capabilities(provider, model)
@@ -137,11 +183,17 @@ def should_route_capture_to_aux_vision(
     if _explicit_aux_vision_override(cfg):
         return True
 
+    user_declared = _lookup_user_declared_supports_vision(provider, model, cfg)
+    if user_declared is True:
+        return False
+    if user_declared is False:
+        return True
+
     accepts_tool_image = _provider_accepts_multimodal_tool_result(provider, model)
     if accepts_tool_image is None or accepts_tool_image is False:
         return True
 
-    supports_vision = _lookup_supports_vision(provider, model)
+    supports_vision = _lookup_supports_vision(provider, model, cfg)
     if supports_vision is True:
         return False
     return True