From 374ba05fe2a4022d95ed45322a13e9eef3ed59de Mon Sep 17 00:00:00 2001
From: tanaka takashi <isazaka@gmail.com>
Date: Wed, 24 Jun 2026 01:51:37 +0900
Subject: vision対応モデル向けに添付画像の処理を変更
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 src/ollama_web/tools/__init__.py | 21 +++++++++++-
 src/ollama_web/tools/image.py    | 72 ++++++++++++++++++++++++++++++++++++++++
 2 files changed, 92 insertions(+), 1 deletion(-)
 create mode 100644 src/ollama_web/tools/image.py

(limited to 'src/ollama_web/tools')

diff --git a/src/ollama_web/tools/__init__.py b/src/ollama_web/tools/__init__.py
index c034c94..17ac505 100644
--- a/src/ollama_web/tools/__init__.py
+++ b/src/ollama_web/tools/__init__.py
@@ -1 +1,20 @@
-"""Tool package for ollama-web."""
\ No newline at end of file
+"""Tool package for ollama-web."""
+
+from .fetch import search_and_fetch
+from .image import resize_image
+from .pdf import extract_pdf_text, pdf_to_text
+from .registry import ToolRegistry, all_tool_callables, default_registry
+from .scrape import scrape_url
+from .search import web_search
+
+__all__ = [
+    "search_and_fetch",
+    "resize_image",
+    "extract_pdf_text",
+    "pdf_to_text",
+    "ToolRegistry",
+    "all_tool_callables",
+    "default_registry",
+    "scrape_url",
+    "web_search",
+]
diff --git a/src/ollama_web/tools/image.py b/src/ollama_web/tools/image.py
new file mode 100644
index 0000000..8bc14f5
--- /dev/null
+++ b/src/ollama_web/tools/image.py
@@ -0,0 +1,72 @@
+"""Image preprocessing helpers for ollama vision models."""
+
+from __future__ import annotations
+
+import io
+from pathlib import Path
+
+from PIL import Image
+
+
+_IMAGE_FORMATS: dict[str, str] = {
+    "png": "PNG",
+    "jpg": "JPEG",
+    "jpeg": "JPEG",
+    "gif": "GIF",
+    "bmp": "BMP",
+    "webp": "WEBP",
+    "tif": "TIFF",
+    "tiff": "TIFF",
+}
+
+
+def _format_from_name(name: str) -> str:
+    ext = Path(name).suffix.lstrip(".").lower()
+    return _IMAGE_FORMATS.get(ext, "JPEG")
+
+
+def resize_image(
+    data: bytes,
+    name: str = "image.jpg",
+    max_dimension: int = 1024,
+    quality: int = 85,
+) -> bytes:
+    """Resize and re-encode an image to reduce payload size for ollama.
+
+    Args:
+      data: Raw image bytes.
+      name: Original file name, used to guess the output format.
+      max_dimension: Maximum width or height in pixels.
+      quality: JPEG/WebP quality when re-encoding to a lossy format.
+
+    Returns:
+      Resized image bytes. If the input is already small, it may be returned
+      unchanged when re-encoding is not needed.
+    """
+    if not data:
+        return data
+
+    try:
+        with Image.open(io.BytesIO(data)) as img:
+            # Preserve orientation if EXIF data exists.
+            img = img.convert("RGB")
+
+            width, height = img.size
+            if width > max_dimension or height > max_dimension:
+                ratio = min(max_dimension / width, max_dimension / height)
+                new_size = (int(width * ratio), int(height * ratio))
+                img = img.resize(new_size, Image.Resampling.LANCZOS)
+
+            fmt = _format_from_name(name)
+            output = io.BytesIO()
+            save_kwargs: dict[str, object] = {}
+            if fmt == "JPEG":
+                save_kwargs["quality"] = quality
+                save_kwargs["optimize"] = True
+            elif fmt == "WEBP":
+                save_kwargs["quality"] = quality
+            img.save(output, format=fmt, **save_kwargs)
+            return output.getvalue()
+    except Exception:  # noqa: BLE001
+        # Fall back to the original bytes if anything goes wrong.
+        return data
\ No newline at end of file
-- 
cgit v1.2.3