機能追加(セッション、左ペインでのセッション管理、ファイル取得ツール、pdfテキスト化ツール)

11 files changed, 1251 insertions, 53 deletions

diff --git a/src/ollama_web/app.py b/src/ollama_web/app.py
index d37557d..7c002b4 100644
--- a/src/ollama_web/app.py
+++ b/src/ollama_web/app.py
@@ -14,6 +14,8 @@ from .config import settings
 from .routes import chat as chat_route
 from .routes import models as models_route
 from .routes import pages as pages_route
+from .routes import sessions as sessions_route
+from .sessions import SessionStore
 
 _STATIC_DIR = Path(__file__).resolve().parent / "static"
 
@@ -25,6 +27,37 @@ def create_app() -> Starlette:
         Route("/api/chat", chat_route.chat, methods=["POST"], name="chat"),
         Route("/api/models", models_route.get_models, methods=["GET"], name="models"),
         Route("/api/health", models_route.health, methods=["GET"], name="health"),
+        Route("/api/sessions", sessions_route.list_sessions, methods=["GET"], name="list_sessions"),
+        Route(
+            "/api/sessions",
+            sessions_route.create_session,
+            methods=["POST"],
+            name="create_session",
+        ),
+        Route(
+            "/api/sessions/{session_id}",
+            sessions_route.get_session,
+            methods=["GET"],
+            name="get_session",
+        ),
+        Route(
+            "/api/sessions/{session_id}",
+            sessions_route.delete_session,
+            methods=["DELETE"],
+            name="delete_session",
+        ),
+        Route(
+            "/api/sessions/{session_id}/files",
+            sessions_route.upload_file,
+            methods=["POST"],
+            name="upload_file",
+        ),
+        Route(
+            "/api/sessions/{session_id}/files/{file_id}",
+            sessions_route.delete_file,
+            methods=["DELETE"],
+            name="delete_file",
+        ),
         Mount("/static", app=StaticFiles(directory=str(_STATIC_DIR)), name="static"),
     ]
 
@@ -39,7 +72,8 @@ def create_app() -> Starlette:
 
     app = Starlette(routes=routes, middleware=middleware)
     app.state.settings = settings
-    app.state.tool_names = ["web_search", "scrape_url"]
+    app.state.session_store = SessionStore(settings.data_dir)
+    app.state.tool_names = ["web_search", "scrape_url", "search_and_fetch", "pdf_to_text"]
     return app
 
 
diff --git a/src/ollama_web/config.py b/src/ollama_web/config.py
index 4f39d17..44947dd 100644
--- a/src/ollama_web/config.py
+++ b/src/ollama_web/config.py
@@ -16,10 +16,18 @@ class Settings:
     )
     host: str = field(default_factory=lambda: os.environ.get("OLLAMA_WEB_HOST", "0.0.0.0"))
     port: int = field(default_factory=lambda: int(os.environ.get("OLLAMA_WEB_PORT", "8000")))
+    data_dir: str = field(
+        default_factory=lambda: os.environ.get("OLLAMA_WEB_DATA_DIR", "data")
+    )
+    max_upload_mb: int = field(
+        default_factory=lambda: int(os.environ.get("OLLAMA_WEB_MAX_UPLOAD_MB", "20"))
+    )
     ollama_timeout: float = 120.0
     scrape_timeout: float = 15.0
     scrape_max_chars: int = 4000
     search_max_results: int = 5
+    fetch_max_results: int = 3
+    fetch_max_chars: int = 8000
 
 
 settings = Settings()
diff --git a/src/ollama_web/routes/chat.py b/src/ollama_web/routes/chat.py
index 2612279..8b6c302 100644
--- a/src/ollama_web/routes/chat.py
+++ b/src/ollama_web/routes/chat.py
@@ -2,7 +2,10 @@
 
 from __future__ import annotations
 
+import functools
+import inspect
 import json
+import logging
 from collections.abc import AsyncIterator
 from typing import Any
 
@@ -10,6 +13,12 @@ from ollama import Message
 from starlette.requests import Request
 from starlette.responses import StreamingResponse
 
+from ..sessions import SessionStore
+from ..tools.fetch import pop_fetched_files
+from ..tools.registry import ToolRegistry, default_registry
+
+logger = logging.getLogger("ollama_web.chat")
+
 
 def _parse_messages(raw: list[dict[str, Any]]) -> list[Message]:
     """Convert raw JSON message dicts into ``ollama.Message`` objects."""
@@ -28,27 +37,127 @@ def _parse_messages(raw: list[dict[str, Any]]) -> list[Message]:
     return out
 
 
-async def _event_stream(request: Request, payload: dict[str, Any]) -> AsyncIterator[bytes]:
+def _session_aware_registry(session_id: str | None) -> ToolRegistry:
+    """Return a registry whose search_and_fetch receives the session id."""
+    reg = default_registry()
+
+    original = reg.get("search_and_fetch")
+    if original is not None and session_id is not None:
+
+        @functools.wraps(original)
+        def wrapped(*args: Any, **kwargs: Any) -> Any:
+            sig = inspect.signature(original)
+            bound = sig.bind(*args, **kwargs)
+            bound.arguments["session_id"] = session_id
+            return original(*bound.args, **bound.kwargs)
+
+        reg._tools["search_and_fetch"] = wrapped  # noqa: SLF001
+
+    return reg
+
+
+def _attachment_context(store: SessionStore, session_id: str, file_ids: list[str]) -> str:
+    parts: list[str] = []
+    for file_id in file_ids:
+        chat_file = store.get_file(session_id, file_id)
+        if chat_file is None:
+            continue
+        text = chat_file.text or ""
+        parts.append(
+            f"<attached-file name=\"{chat_file.name}\" id=\"{chat_file.id}\" "
+            f"source=\"{chat_file.source}\">\n{text}\n</attached-file>"
+        )
+    if not parts:
+        return ""
+    return "\n\n".join(["以下はユーザーが添付したファイルの内容です：", *parts])
+
+
+async def _event_stream(
+    request: Request,
+    payload: dict[str, Any],
+    session_id: str | None,
+) -> AsyncIterator[bytes]:
     """Yield SSE events from the tool-augmented chat loop."""
     model = payload.get("model") or request.app.state.settings.default_model
-    raw_messages = payload.get("messages", [])
     think = payload.get("think")
 
+    store: SessionStore = request.app.state.session_store
+    file_ids: list[str] = list(payload.get("file_ids", []))
+
+    session_messages: list[Message] = []
+    if session_id:
+        session = store.get(session_id)
+        if session is not None:
+            session_messages = _parse_messages(session.get("messages", []))
+
+    user_content = ""
+    raw_messages = payload.get("messages", [])
+    if raw_messages:
+        user_content = str(raw_messages[-1].get("content", "") or "")
+
+    attachment_ctx = ""
+    if session_id and file_ids:
+        attachment_ctx = _attachment_context(store, session_id, file_ids)
+
+    if attachment_ctx and user_content:
+        # Inject attachment context into the latest user message.
+        augmented = f"{user_content}\n\n{attachment_ctx}"
+        if raw_messages:
+            raw_messages[-1]["content"] = augmented
+    elif attachment_ctx:
+        raw_messages.append({"role": "user", "content": attachment_ctx})
+
     messages = _parse_messages(raw_messages)
+    if session_messages:
+        messages = session_messages + messages
+
     host = request.app.state.settings.ollama_host
 
     from ollama_web import llm  # local import to avoid circular at module load
 
-    async for event in llm.astream_chat_with_tools(messages, model, think=think, host=host):
+    registry = _session_aware_registry(session_id)
+
+    assistant_text = ""
+    finished = False
+    async for event in llm.astream_chat_with_tools(messages, model, think=think, host=host, registry=registry):
+        if event.get("type") == "delta":
+            assistant_text += event.get("content", "")
+        elif event.get("type") == "done":
+            finished = True
         data = json.dumps(event, ensure_ascii=False)
         yield f"data: {data}\n\n".encode()
 
+    # Persist user and assistant messages to the session.
+    if session_id and user_content:
+        store.add_message(
+            session_id=session_id,
+            role="user",
+            content=user_content,
+            attachments=file_ids,
+        )
+        if assistant_text:
+            store.add_message(
+                session_id=session_id,
+                role="assistant",
+                content=assistant_text,
+            )
+
+    # Persist fetched files from tools like search_and_fetch.
+    if session_id:
+        fetched = pop_fetched_files(session_id)
+        for f in fetched:
+            store.add_file(
+                session_id=session_id,
+                name=f["name"],
+                data=f["data"],
+                mime=f["mime"],
+                text=f["text"],
+                source="fetch",
+            )
+
 
 async def chat(request: Request) -> StreamingResponse:
     """POST /api/chat: stream chat events as Server-Sent Events."""
-    import logging
-
-    logger = logging.getLogger("ollama_web.chat")
     try:
         body = await request.body()
         payload = json.loads(body.decode("utf-8"))
@@ -59,14 +168,16 @@ async def chat(request: Request) -> StreamingResponse:
             _single_event(data),
             media_type="text/event-stream",
         )
+
+    session_id = payload.get("session_id")
     msg_count = len(payload.get("messages", []))
-    logger.info("chat request model=%s messages=%d", payload.get("model"), msg_count)
+    logger.info("chat request model=%s session=%s messages=%d", payload.get("model"), session_id, msg_count)
     return StreamingResponse(
-        _event_stream(request, payload),
+        _event_stream(request, payload, session_id),
         media_type="text/event-stream",
     )
 
 
 async def _single_event(data: str) -> AsyncIterator[bytes]:
     """Yield a single SSE event and finish."""
-    yield f"data: {data}\n\n".encode()
+    yield f"data: {data}\n\n".encode()
\ No newline at end of file
diff --git a/src/ollama_web/routes/sessions.py b/src/ollama_web/routes/sessions.py
new file mode 100644
index 0000000..8980960
--- /dev/null
+++ b/src/ollama_web/routes/sessions.py
@@ -0,0 +1,120 @@
+"""Session management API routes."""
+
+from __future__ import annotations
+
+import logging
+from typing import Any
+
+from starlette.requests import Request
+from starlette.responses import JSONResponse
+
+from ..sessions import SessionStore
+from ..tools.pdf import extract_pdf_text
+
+logger = logging.getLogger("ollama_web.sessions")
+
+
+def _get_store(request: Request) -> SessionStore:
+    return request.app.state.session_store
+
+
+async def list_sessions(request: Request) -> JSONResponse:
+    store = _get_store(request)
+    return JSONResponse({"sessions": store.list_sessions()})
+
+
+async def create_session(request: Request) -> JSONResponse:
+    store = _get_store(request)
+    body = await request.json() if await request.body() else {}
+    title = body.get("title")
+    session = store.create(title=title)
+    return JSONResponse(session)
+
+
+async def get_session(request: Request) -> JSONResponse:
+    store = _get_store(request)
+    session_id = request.path_params["session_id"]
+    session = store.get(session_id)
+    if session is None:
+        return JSONResponse({"error": "Session not found"}, status_code=404)
+    return JSONResponse(session)
+
+
+async def delete_session(request: Request) -> JSONResponse:
+    store = _get_store(request)
+    session_id = request.path_params["session_id"]
+    ok = store.delete(session_id)
+    if not ok:
+        return JSONResponse({"error": "Session not found"}, status_code=404)
+    return JSONResponse({"deleted": True})
+
+
+async def upload_file(request: Request) -> JSONResponse:
+    store = _get_store(request)
+    session_id = request.path_params["session_id"]
+    session = store.get(session_id)
+    if session is None:
+        return JSONResponse({"error": "Session not found"}, status_code=404)
+
+    max_size = request.app.state.settings.max_upload_mb * 1024 * 1024
+    try:
+        form = await request.form()
+    except Exception as exc:  # noqa: BLE001
+        logger.error("failed to parse multipart form: %s", exc)
+        return JSONResponse({"error": f"Invalid form data: {exc}"}, status_code=400)
+
+    files = form.getlist("files") if hasattr(form, "getlist") else [form.get("files")]
+    results: list[dict[str, Any]] = []
+    for upload in files:
+        if upload is None:
+            continue
+        try:
+            data = await upload.read()
+        except Exception as exc:  # noqa: BLE001
+            results.append({"name": getattr(upload, "filename", "unknown"), "error": str(exc)})
+            continue
+        if len(data) > max_size:
+            results.append(
+                {
+                    "name": getattr(upload, "filename", "unknown"),
+                    "error": f"File exceeds {request.app.state.settings.max_upload_mb} MB limit",
+                }
+            )
+            continue
+
+        name = getattr(upload, "filename", "uploaded-file") or "uploaded-file"
+        content_type = getattr(upload, "content_type", None) or "application/octet-stream"
+        text = ""
+        if content_type == "application/pdf" or name.lower().endswith(".pdf"):
+            text = extract_pdf_text(data)
+        else:
+            # Try decode as text for text-like uploads.
+            try:
+                text = data.decode("utf-8")
+            except UnicodeDecodeError:
+                try:
+                    text = data.decode("cp932")
+                except UnicodeDecodeError:
+                    text = data.decode("utf-8", errors="replace")
+
+        chat_file = store.add_file(
+            session_id=session_id,
+            name=name,
+            data=data,
+            mime=content_type,
+            text=text,
+            source="upload",
+        )
+        results.append(chat_file.to_dict())
+
+    return JSONResponse({"files": results})
+
+
+async def delete_file(request: Request) -> JSONResponse:
+    store = _get_store(request)
+    session_id = request.path_params["session_id"]
+    file_id = request.path_params["file_id"]
+    ok = store.remove_file(session_id, file_id)
+    if not ok:
+        return JSONResponse({"error": "File not found"}, status_code=404)
+    return JSONResponse({"deleted": True})
\ No newline at end of file
diff --git a/src/ollama_web/sessions.py b/src/ollama_web/sessions.py
new file mode 100644
index 0000000..bfe659c
--- /dev/null
+++ b/src/ollama_web/sessions.py
@@ -0,0 +1,248 @@
+"""Persistent session storage for ollama-web."""
+
+from __future__ import annotations
+
+import json
+import shutil
+import uuid
+from dataclasses import dataclass, field
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any
+
+from .config import settings
+
+
+def _now() -> str:
+    return datetime.now(timezone.utc).isoformat()
+
+
+@dataclass
+class ChatFile:
+    id: str
+    name: str
+    path: str
+    mime: str
+    size: int
+    text: str
+    source: str  # "upload" | "fetch"
+
+    def to_dict(self) -> dict[str, Any]:
+        return {
+            "id": self.id,
+            "name": self.name,
+            "path": self.path,
+            "mime": self.mime,
+            "size": self.size,
+            "text": self.text,
+            "source": self.source,
+        }
+
+    @classmethod
+    def from_dict(cls, data: dict[str, Any]) -> "ChatFile":
+        return cls(
+            id=data["id"],
+            name=data["name"],
+            path=data["path"],
+            mime=data.get("mime", "application/octet-stream"),
+            size=data.get("size", 0),
+            text=data.get("text", ""),
+            source=data.get("source", "upload"),
+        )
+
+
+@dataclass
+class ChatMessage:
+    role: str
+    content: str
+    attachments: list[str] = field(default_factory=list)
+
+    def to_dict(self) -> dict[str, Any]:
+        return {
+            "role": self.role,
+            "content": self.content,
+            "attachments": list(self.attachments),
+        }
+
+    @classmethod
+    def from_dict(cls, data: dict[str, Any]) -> "ChatMessage":
+        return cls(
+            role=data.get("role", "user"),
+            content=data.get("content", ""),
+            attachments=list(data.get("attachments", [])),
+        )
+
+
+class SessionStore:
+    """Manage chat sessions as JSON files under the configured data directory."""
+
+    def __init__(self, data_dir: str | Path | None = None) -> None:
+        self.data_dir = Path(data_dir or settings.data_dir).resolve()
+        self.sessions_dir = self.data_dir / "sessions"
+        self.sessions_dir.mkdir(parents=True, exist_ok=True)
+
+    def _session_dir(self, session_id: str) -> Path:
+        return self.sessions_dir / session_id
+
+    def _session_file(self, session_id: str) -> Path:
+        return self._session_dir(session_id) / "session.json"
+
+    def _files_dir(self, session_id: str) -> Path:
+        files_dir = self._session_dir(session_id) / "files"
+        files_dir.mkdir(parents=True, exist_ok=True)
+        return files_dir
+
+    def list_sessions(self) -> list[dict[str, Any]]:
+        out: list[dict[str, Any]] = []
+        if not self.sessions_dir.exists():
+            return out
+        for entry in sorted(self.sessions_dir.iterdir(), key=lambda p: p.stat().st_mtime, reverse=True):
+            if not entry.is_dir():
+                continue
+            session_file = entry / "session.json"
+            if not session_file.exists():
+                continue
+            try:
+                data = json.loads(session_file.read_text(encoding="utf-8"))
+                out.append(
+                    {
+                        "id": data["id"],
+                        "title": data.get("title", "Untitled"),
+                        "updated_at": data.get("updated_at", data.get("created_at", "")),
+                    }
+                )
+            except Exception:  # noqa: BLE001
+                continue
+        return out
+
+    def create(self, title: str | None = None) -> dict[str, Any]:
+        session_id = uuid.uuid4().hex
+        now = _now()
+        session = {
+            "id": session_id,
+            "title": title or "新しいセッション",
+            "created_at": now,
+            "updated_at": now,
+            "messages": [],
+            "files": [],
+        }
+        self._session_dir(session_id).mkdir(parents=True, exist_ok=True)
+        self._save(session_id, session)
+        return session
+
+    def get(self, session_id: str) -> dict[str, Any] | None:
+        session_file = self._session_file(session_id)
+        if not session_file.exists():
+            return None
+        try:
+            return json.loads(session_file.read_text(encoding="utf-8"))
+        except Exception:  # noqa: BLE001
+            return None
+
+    def save(self, session: dict[str, Any]) -> None:
+        session["updated_at"] = _now()
+        self._save(session["id"], session)
+
+    def _save(self, session_id: str, session: dict[str, Any]) -> None:
+        session_file = self._session_file(session_id)
+        session_file.parent.mkdir(parents=True, exist_ok=True)
+        session_file.write_text(json.dumps(session, ensure_ascii=False, indent=2), encoding="utf-8")
+
+    def delete(self, session_id: str) -> bool:
+        session_dir = self._session_dir(session_id)
+        if not session_dir.exists():
+            return False
+        try:
+            shutil.rmtree(session_dir)
+            return True
+        except Exception:  # noqa: BLE001
+            return False
+
+    def add_message(
+        self,
+        session_id: str,
+        role: str,
+        content: str,
+        attachments: list[str] | None = None,
+    ) -> dict[str, Any] | None:
+        session = self.get(session_id)
+        if session is None:
+            return None
+        session["messages"].append(
+            ChatMessage(
+                role=role,
+                content=content,
+                attachments=list(attachments or []),
+            ).to_dict()
+        )
+        if role == "user" and len(session["messages"]) <= 2:
+            first_user = next(
+                (m for m in session["messages"] if m["role"] == "user"),
+                None,
+            )
+            if first_user:
+                title = first_user["content"].strip().replace("\n", " ")[:40]
+                session["title"] = title + ("…" if len(first_user["content"]) > 40 else "")
+        self.save(session)
+        return session
+
+    def add_file(
+        self,
+        session_id: str,
+        name: str,
+        data: bytes,
+        mime: str,
+        text: str,
+        source: str = "upload",
+    ) -> ChatFile:
+        self._session_dir(session_id).mkdir(parents=True, exist_ok=True)
+        file_id = uuid.uuid4().hex
+        ext = Path(name).suffix
+        stored_name = f"{file_id}{ext}"
+        file_path = self._files_dir(session_id) / stored_name
+        file_path.write_bytes(data)
+        rel_path = str((self._session_dir(session_id) / "files" / stored_name).relative_to(self.data_dir))
+        chat_file = ChatFile(
+            id=file_id,
+            name=name,
+            path=rel_path,
+            mime=mime,
+            size=len(data),
+            text=text,
+            source=source,
+        )
+        session = self.get(session_id)
+        if session is None:
+            session = self.create()
+            session_id = session["id"]
+        session["files"].append(chat_file.to_dict())
+        self.save(session)
+        return chat_file
+
+    def remove_file(self, session_id: str, file_id: str) -> bool:
+        session = self.get(session_id)
+        if session is None:
+            return False
+        files = session.get("files", [])
+        target = next((f for f in files if f["id"] == file_id), None)
+        if target is None:
+            return False
+        files.remove(target)
+        session["files"] = files
+        try:
+            abs_path = self.data_dir / target["path"]
+            if abs_path.exists():
+                abs_path.unlink()
+        except Exception:  # noqa: BLE001
+            pass
+        self.save(session)
+        return True
+
+    def get_file(self, session_id: str, file_id: str) -> ChatFile | None:
+        session = self.get(session_id)
+        if session is None:
+            return None
+        for f in session.get("files", []):
+            if f["id"] == file_id:
+                return ChatFile.from_dict(f)
+        return None
\ No newline at end of file
diff --git a/src/ollama_web/static/app.js b/src/ollama_web/static/app.js
index 845a71f..339030a 100644
--- a/src/ollama_web/static/app.js
+++ b/src/ollama_web/static/app.js
@@ -8,12 +8,17 @@ const modelSelect = document.getElementById("model-select");
 const thinkToggle = document.getElementById("think-toggle");
 const clearBtn = document.getElementById("clear-btn");
 const cancelBtn = document.getElementById("cancel-btn");
+const fileInput = document.getElementById("file-input");
+const attachmentsEl = document.getElementById("attachments");
+const sessionListEl = document.getElementById("session-list");
+const newSessionBtn = document.getElementById("new-session-btn");
 
-let messages = [];
+let currentSessionId = null;
+let currentSession = null;
+let pendingFiles = [];
 let abortController = null;
 
 function simpleMarkdownToHtml(text) {
-  // Render plain text safely as HTML without external dependencies.
   const el = document.createElement("div");
   el.setAttribute("style", "white-space:pre-wrap");
   el.textContent = text;
@@ -115,19 +120,211 @@ function parseSseEvent(chunk) {
   return events;
 }
 
+async function loadSessions() {
+  try {
+    const res = await fetch("/api/sessions");
+    if (!res.ok) throw new Error(res.statusText);
+    const data = await res.json();
+    renderSessionList(data.sessions || []);
+  } catch (err) {
+    console.error("failed to load sessions", err);
+    renderSessionList([]);
+  }
+}
+
+function renderSessionList(sessions) {
+  sessionListEl.innerHTML = "";
+  if (!sessions.length) {
+    const li = document.createElement("li");
+    li.className = "empty";
+    li.textContent = "セッションがありません";
+    sessionListEl.appendChild(li);
+    return;
+  }
+  for (const s of sessions) {
+    const li = document.createElement("li");
+    li.dataset.id = s.id;
+    if (s.id === currentSessionId) li.classList.add("active");
+    const title = document.createElement("span");
+    title.className = "title";
+    title.textContent = s.title || s.id;
+    const del = document.createElement("button");
+    del.className = "delete";
+    del.textContent = "削除";
+    del.title = "セッションを削除";
+    del.addEventListener("click", (e) => {
+      e.stopPropagation();
+      deleteSession(s.id);
+    });
+    li.append(title, del);
+    li.addEventListener("click", () => selectSession(s.id));
+    sessionListEl.appendChild(li);
+  }
+}
+
+async function createSession() {
+  try {
+    const res = await fetch("/api/sessions", { method: "POST" });
+    if (!res.ok) throw new Error(res.statusText);
+    const session = await res.json();
+    await loadSessions();
+    await selectSession(session.id);
+  } catch (err) {
+    console.error("failed to create session", err);
+    alert("セッションの作成に失敗しました");
+  }
+}
+
+async function deleteSession(id) {
+  if (!confirm("このセッションを削除しますか？")) return;
+  try {
+    const res = await fetch(`/api/sessions/${id}`, { method: "DELETE" });
+    if (!res.ok) throw new Error(res.statusText);
+    if (currentSessionId === id) {
+      currentSessionId = null;
+      currentSession = null;
+      chatEl.innerHTML = "";
+      pendingFiles = [];
+      renderAttachments();
+    }
+    await loadSessions();
+  } catch (err) {
+    console.error("failed to delete session", err);
+    alert("セッションの削除に失敗しました");
+  }
+}
+
+async function selectSession(id) {
+  try {
+    const res = await fetch(`/api/sessions/${id}`);
+    if (!res.ok) throw new Error(res.statusText);
+    const session = await res.json();
+    currentSessionId = session.id;
+    currentSession = session;
+    renderSession();
+    renderSessionList((await (await fetch("/api/sessions")).json()).sessions || []);
+  } catch (err) {
+    console.error("failed to select session", err);
+  }
+}
+
+function renderSession() {
+  chatEl.innerHTML = "";
+  if (!currentSession) return;
+  const msgs = currentSession.messages || [];
+  for (const m of msgs) {
+    if (m.role === "user") {
+      const body = addMessage("user").body;
+      body.innerHTML = renderMarkdown(m.content);
+      highlightIn(body);
+      if (m.attachments && m.attachments.length) {
+        const info = document.createElement("div");
+        info.style.fontSize = "12px";
+        info.style.color = "var(--muted)";
+        info.style.marginTop = "6px";
+        const names = m.attachments
+          .map((fid) => {
+            const f = (currentSession.files || []).find((x) => x.id === fid);
+            return f ? f.name : fid;
+          })
+          .join(", ");
+        info.textContent = `添付: ${names}`;
+        body.appendChild(info);
+      }
+    } else if (m.role === "assistant") {
+      const body = addMessage("assistant").body;
+      body.innerHTML = renderMarkdown(m.content);
+      highlightIn(body);
+    }
+  }
+}
+
+function renderAttachments() {
+  attachmentsEl.innerHTML = "";
+  for (const f of pendingFiles) {
+    const chip = document.createElement("span");
+    chip.className = "attachment-chip";
+    chip.textContent = f.name;
+    const remove = document.createElement("span");
+    remove.className = "remove";
+    remove.textContent = "×";
+    remove.addEventListener("click", () => {
+      pendingFiles = pendingFiles.filter((x) => x !== f);
+      renderAttachments();
+    });
+    chip.appendChild(remove);
+    attachmentsEl.appendChild(chip);
+  }
+}
+
+async function uploadFiles(files) {
+  if (!currentSessionId) {
+    await createSession();
+  }
+  if (!currentSessionId) return;
+
+  const form = new FormData();
+  for (const file of files) {
+    form.append("files", file);
+  }
+  try {
+    const res = await fetch(`/api/sessions/${currentSessionId}/files`, {
+      method: "POST",
+      body: form,
+    });
+    if (!res.ok) throw new Error(res.statusText);
+    const data = await res.json();
+    for (const f of data.files || []) {
+      if (f.error) {
+        alert(`${f.name}: ${f.error}`);
+        continue;
+      }
+      pendingFiles.push({ id: f.id, name: f.name });
+    }
+    renderAttachments();
+    await selectSession(currentSessionId);
+  } catch (err) {
+    console.error("failed to upload files", err);
+    alert("ファイルのアップロードに失敗しました");
+  }
+}
+
+fileInput.addEventListener("change", () => {
+  if (fileInput.files && fileInput.files.length) {
+    uploadFiles(fileInput.files);
+    fileInput.value = "";
+  }
+});
+
+newSessionBtn.addEventListener("click", createSession);
+
 async function send() {
   const text = inputEl.value.trim();
-  if (!text) return;
-  inputEl.value = "";
+  if (!text && !pendingFiles.length) return;
+
+  if (!currentSessionId) {
+    await createSession();
+  }
+  if (!currentSessionId) return;
+
+  if (text) inputEl.value = "";
   sendBtn.disabled = true;
   cancelBtn.disabled = false;
   abortController = new AbortController();
 
-  messages.push({ role: "user", content: text });
+  const fileIds = pendingFiles.map((f) => f.id);
   const userBody = addMessage("user").body;
-  userBody.innerHTML = renderMarkdown(text);
+  let displayContent = text;
+  if (pendingFiles.length) {
+    const names = pendingFiles.map((f) => f.name).join(", ");
+    displayContent += (text ? "\n\n" : "") + `添付ファイル: ${names}`;
+  }
+  userBody.innerHTML = renderMarkdown(displayContent);
   highlightIn(userBody);
 
+  pendingFiles = [];
+  renderAttachments();
+
   const assistantMsg = addMessage("assistant");
   const assistantBody = assistantMsg.body;
   const assistantExtras = assistantMsg.extras;
@@ -139,7 +336,14 @@ async function send() {
   const model = modelSelect.value;
   const think = thinkToggle.checked;
 
-  const requestBody = JSON.stringify({ model, messages, think });
+  const requestBody = JSON.stringify({
+    model,
+    messages: text ? [{ role: "user", content: text }] : [],
+    think,
+    session_id: currentSessionId,
+    file_ids: fileIds,
+  });
+
   let res;
   try {
     res = await fetch("/api/chat", {
@@ -169,7 +373,6 @@ async function send() {
   const decoder = new TextDecoder();
   let buffer = "";
 
-  // Remove the placeholder "…" as soon as real content arrives.
   let placeholderCleared = false;
 
   function appendDelta(content) {
@@ -188,7 +391,7 @@ async function send() {
     if (done) break;
     buffer += decoder.decode(value, { stream: true });
 
-    const endIndex = buffer.lastIndexOf("\n\n");
+    const endIndex = buffer.lastIndexOf("\\n\\n");
     if (endIndex === -1) continue;
     const chunk = buffer.slice(0, endIndex);
     buffer = buffer.slice(endIndex + 2);
@@ -250,13 +453,12 @@ async function send() {
     assistantBody.textContent = "（応答なし）";
   }
 
-  if (assistantText) {
-    messages.push({ role: "assistant", content: assistantText });
-  }
   abortController = null;
   cancelBtn.disabled = true;
   sendBtn.disabled = false;
   inputEl.focus();
+
+  await selectSession(currentSessionId);
 }
 
 formEl.addEventListener("submit", (e) => {
@@ -272,7 +474,6 @@ inputEl.addEventListener("keydown", (e) => {
 });
 
 clearBtn.addEventListener("click", () => {
-  messages = [];
   chatEl.innerHTML = "";
 });
 
@@ -280,4 +481,6 @@ cancelBtn.addEventListener("click", () => {
   if (abortController) {
     abortController.abort();
   }
-});
\ No newline at end of file
+});
+
+loadSessions();
\ No newline at end of file
diff --git a/src/ollama_web/static/style.css b/src/ollama_web/static/style.css
index 0be42c2..d12cf48 100644
--- a/src/ollama_web/static/style.css
+++ b/src/ollama_web/static/style.css
@@ -9,6 +9,7 @@
   --assistant-bg: #161b22;
   --tool-bg: #2d1117;
   --think-bg: #1a2332;
+  --danger: #f85149;
 }
 
 * { box-sizing: border-box; }
@@ -24,8 +25,112 @@ html, body {
 
 body {
   display: flex;
-  flex-direction: column;
   height: 100vh;
+  overflow: hidden;
+}
+
+#sidebar {
+  width: 260px;
+  min-width: 260px;
+  background: var(--panel);
+  border-right: 1px solid var(--border);
+  display: flex;
+  flex-direction: column;
+}
+
+.sidebar-header {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  padding: 12px 14px;
+  border-bottom: 1px solid var(--border);
+}
+
+.sidebar-header h2 {
+  font-size: 15px;
+  margin: 0;
+  color: var(--muted);
+}
+
+.sidebar-header button {
+  background: transparent;
+  border: 1px solid var(--border);
+  border-radius: 6px;
+  color: var(--accent);
+  cursor: pointer;
+  font-size: 18px;
+  line-height: 1;
+  padding: 2px 8px;
+}
+
+.sidebar-header button:hover {
+  border-color: var(--accent);
+}
+
+#session-list {
+  list-style: none;
+  margin: 0;
+  padding: 8px;
+  overflow-y: auto;
+  flex: 1;
+}
+
+#session-list li {
+  display: flex;
+  align-items: center;
+  gap: 6px;
+  padding: 8px 10px;
+  border-radius: 8px;
+  cursor: pointer;
+  margin-bottom: 4px;
+}
+
+#session-list li:hover {
+  background: #21262d;
+}
+
+#session-list li.active {
+  background: #1f6feb33;
+  border: 1px solid var(--accent);
+}
+
+#session-list li .title {
+  flex: 1;
+  white-space: nowrap;
+  overflow: hidden;
+  text-overflow: ellipsis;
+  font-size: 14px;
+}
+
+#session-list li .delete {
+  display: none;
+  background: transparent;
+  border: none;
+  color: var(--danger);
+  cursor: pointer;
+  font-size: 12px;
+  padding: 2px 4px;
+}
+
+#session-list li:hover .delete {
+  display: inline;
+}
+
+#session-list li.empty {
+  color: var(--muted);
+  font-size: 13px;
+  cursor: default;
+}
+
+#session-list li.empty:hover {
+  background: transparent;
+}
+
+#main {
+  flex: 1;
+  display: flex;
+  flex-direction: column;
+  min-width: 0;
 }
 
 header {
@@ -134,14 +239,67 @@ main#chat {
   background: var(--think-bg);
 }
 
-form#input-form {
+#attachments {
   display: flex;
-  gap: 8px;
+  flex-wrap: wrap;
+  gap: 6px;
+  padding: 6px 16px 0;
+  background: var(--panel);
+  border-top: 1px solid var(--border);
+  min-height: 0;
+}
+
+#attachments:empty {
+  display: none;
+}
+
+.attachment-chip {
+  display: inline-flex;
+  align-items: center;
+  gap: 6px;
+  background: var(--bg);
+  border: 1px solid var(--border);
+  border-radius: 6px;
+  padding: 4px 8px;
+  font-size: 13px;
+}
+
+.attachment-chip .remove {
+  cursor: pointer;
+  color: var(--danger);
+  font-weight: 600;
+}
+
+form#input-form {
   padding: 12px 16px;
   background: var(--panel);
   border-top: 1px solid var(--border);
 }
 
+.input-row {
+  display: flex;
+  gap: 8px;
+  align-items: stretch;
+}
+
+.attach-btn {
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  width: 42px;
+  background: var(--bg);
+  border: 1px solid var(--border);
+  border-radius: 8px;
+  cursor: pointer;
+  font-size: 20px;
+}
+
+.attach-btn:hover { border-color: var(--accent); }
+
+.attach-btn input[type="file"] {
+  display: none;
+}
+
 textarea#input {
   flex: 1;
   resize: none;
diff --git a/src/ollama_web/templates/index.html b/src/ollama_web/templates/index.html
index 14bdf87..eae3b6d 100644
--- a/src/ollama_web/templates/index.html
+++ b/src/ollama_web/templates/index.html
@@ -10,41 +10,61 @@
   <script src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/11.9.0/highlight.min.js"></script>
 </head>
 <body>
-  <header>
-    <h1>ollama-web</h1>
-    <div class="controls">
-      <label for="model-select">モデル:</label>
-      <select id="model-select">
-        {% if models %}
-          {% for m in models %}
-            <option value="{{ m }}" {% if m == default_model %}selected{% endif %}>{{ m }}</option>
-          {% endfor %}
-        {% else %}
-          <option value="{{ default_model }}" selected>{{ default_model }}</option>
-        {% endif %}
-      </select>
-      <label class="think-toggle">
-        <input type="checkbox" id="think-toggle" /> thinking
-      </label>
-      <button id="clear-btn" type="button">会話クリア</button>
+  <aside id="sidebar">
+    <div class="sidebar-header">
+      <h2>セッション</h2>
+      <button id="new-session-btn" type="button" title="新規セッション">＋</button>
     </div>
-  </header>
+    <ul id="session-list"></ul>
+  </aside>
 
-  <main id="chat"></main>
+  <div id="main">
+    <header>
+      <h1>ollama-web</h1>
+      <div class="controls">
+        <label for="model-select">モデル:</label>
+        <select id="model-select">
+          {% if models %}
+            {% for m in models %}
+              <option value="{{ m }}" {% if m == default_model %}selected{% endif %}>{{ m }}</option>
+            {% endfor %}
+          {% else %}
+            <option value="{{ default_model }}" selected>{{ default_model }}</option>
+          {% endif %}
+        </select>
+        <label class="think-toggle">
+          <input type="checkbox" id="think-toggle" /> thinking
+        </label>
+        <button id="clear-btn" type="button">会話クリア</button>
+      </div>
+    </header>
 
-  <form id="input-form">
-    <textarea id="input" placeholder="メッセージを入力… (Enter で送信 / Shift+Enter で改行)"
-              rows="2" autofocus></textarea>
-    <button type="submit" id="send-btn">送信</button>
-    <button type="button" id="cancel-btn" disabled>停止</button>
-  </form>
+    <main id="chat"></main>
+
+    <div id="attachments"></div>
+
+    <form id="input-form">
+      <div class="input-row">
+        <label class="attach-btn" title="ファイルを添付">
+          <input type="file" id="file-input" multiple accept="*/*" />
+          📎
+        </label>
+        <textarea id="input" placeholder="メッセージを入力… (Enter で送信 / Shift+Enter で改行)"
+                  rows="2" autofocus></textarea>
+        <button type="submit" id="send-btn">送信</button>
+        <button type="button" id="cancel-btn" disabled>停止</button>
+      </div>
+    </form>
+  </div>
 
   <script>
     const TOOLS = [
       { name: "web_search", label: "Web検索" },
       { name: "scrape_url", label: "スクレイピング" },
+      { name: "search_and_fetch", label: "ファイル検索・取得" },
+      { name: "pdf_to_text", label: "PDFテキスト変換" },
     ];
   </script>
   <script src="{{ url_for('static', path='app.js') }}"></script>
 </body>
-</html>
+</html>
\ No newline at end of file
diff --git a/src/ollama_web/tools/fetch.py b/src/ollama_web/tools/fetch.py
new file mode 100644
index 0000000..62670b2
--- /dev/null
+++ b/src/ollama_web/tools/fetch.py
@@ -0,0 +1,204 @@
+"""Search-and-fetch tool that finds files on the web by type and returns text."""
+
+from __future__ import annotations
+
+import base64
+import mimetypes
+import re
+from pathlib import Path
+from typing import Any
+
+import httpx
+
+from ..config import settings
+from .pdf import extract_pdf_text
+from .scrape import scrape_url
+
+
+_FetchedFile = dict[str, Any]
+
+# Per-session temporary storage for files fetched by search_and_fetch.
+# The chat route drains this after the tool call finishes.
+_fetched_files: dict[str, list[_FetchedFile]] = {}
+
+
+_FILETYPE_SUFFIXES: dict[str, str] = {
+    "pdf": "pdf",
+    "txt": "txt",
+    "csv": "csv",
+    "md": "md",
+    "json": "json",
+    "html": "html",
+    "htm": "htm",
+    "xml": "xml",
+    "py": "py",
+    "js": "js",
+    "ts": "ts",
+}
+
+
+def set_session_context(session_id: str | None) -> None:
+    """Set the active session id so fetched files can be persisted later."""
+    # No-op marker kept for future hooks; storage is keyed by session_id directly.
+    pass
+
+
+def store_fetched_file(session_id: str, name: str, data: bytes, mime: str, text: str, url: str) -> None:
+    """Store a fetched file so the chat route can persist it to the session."""
+    if session_id:
+        _fetched_files.setdefault(session_id, []).append(
+            {
+                "name": name,
+                "data": data,
+                "mime": mime,
+                "text": text,
+                "url": url,
+            }
+        )
+
+
+def pop_fetched_files(session_id: str) -> list[_FetchedFile]:
+    """Return and clear any files fetched for the given session."""
+    return _fetched_files.pop(session_id, [])
+
+
+def _normalize_file_type(file_type: str) -> str:
+    ft = file_type.strip().lower().lstrip(".")
+    return _FILETYPE_SUFFIXES.get(ft, ft)
+
+
+def _filetype_query(query: str, file_type: str) -> str:
+    return f"{query} filetype:{file_type}"
+
+
+def _is_matching_url(url: str, file_type: str) -> bool:
+    suffix = f".{file_type}"
+    parsed = httpx.URL(url)
+    path = parsed.path or ""
+    if path.lower().endswith(suffix):
+        return True
+    # Some CDNs embed extension after query params.
+    if re.search(rf"\.{re.escape(file_type)}(?:[?#]|$)", url, re.IGNORECASE):
+        return True
+    return False
+
+
+def _guess_mime(name: str, file_type: str) -> str:
+    mime, _ = mimetypes.guess_type(name)
+    return mime or f"application/{file_type}"
+
+
+def _fetch_bytes(url: str) -> bytes:
+    with httpx.Client(
+        timeout=settings.scrape_timeout,
+        follow_redirects=True,
+        headers={"User-Agent": "ollama-web/0.1 (+https://github.com/local)"},
+    ) as client:
+        resp = client.get(url)
+        resp.raise_for_status()
+        return resp.content
+
+
+def _text_from_bytes(url: str, data: bytes, file_type: str, max_chars: int) -> str:
+    name = Path(httpx.URL(url).path or "file").name or f"file.{file_type}"
+    if not name or "." not in name:
+        name = f"file.{file_type}"
+
+    if file_type == "pdf":
+        return extract_pdf_text(data, max_chars=max_chars)
+
+    # Decode text-like formats.
+    try:
+        text = data.decode("utf-8")
+    except UnicodeDecodeError:
+        try:
+            text = data.decode("cp932")
+        except UnicodeDecodeError:
+            text = data.decode("utf-8", errors="replace")
+
+    if file_type in {"html", "htm"}:
+        text = scrape_url(url, max_chars=max_chars)
+
+    if len(text) > max_chars:
+        text = text[:max_chars].rsplit(" ", 1)[0] + "\n…[truncated]"
+    return text
+
+
+def search_and_fetch(
+    query: str,
+    file_type: str,
+    max_results: int | None = None,
+    max_chars: int | None = None,
+    session_id: str | None = None,
+) -> str:
+    """Search the web for files of a given type and return their text content.
+
+    Args:
+      query: Search keywords.
+      file_type: Desired file extension (e.g. "pdf", "csv", "txt", "md", "json").
+      max_results: Number of candidate URLs to fetch.
+      max_chars: Max characters per file to return.
+      session_id: If provided, fetched files are queued for session persistence.
+
+    Returns:
+      A formatted string containing metadata and extracted text for each
+      successfully fetched file.
+    """
+    if not query or not query.strip():
+        return "Empty query."
+
+    ft = _normalize_file_type(file_type)
+    if not ft:
+        return "Invalid file type."
+
+    limit = (
+        max_results
+        if max_results is not None and max_results > 0
+        else settings.fetch_max_results
+    )
+    per_file_chars = (
+        max_chars if max_chars is not None and max_chars > 0 else settings.fetch_max_chars
+    )
+
+    from .search import web_search
+
+    search_query = _filetype_query(query, ft)
+    results = web_search(search_query, max_results=limit * 2)
+
+    matched: list[dict[str, str]] = []
+    for item in results:
+        href = item.get("href", "")
+        if _is_matching_url(href, ft):
+            matched.append(item)
+            if len(matched) >= limit:
+                break
+
+    if not matched:
+        return f"No {ft} files found for query: {query}"
+
+    outputs: list[str] = []
+    for idx, item in enumerate(matched, 1):
+        url = item.get("href", "")
+        title = item.get("title", "")
+        try:
+            data = _fetch_bytes(url)
+        except Exception as exc:  # noqa: BLE001
+            outputs.append(f"=== 候補 {idx} ===\nURL: {url}\nタイトル: {title}\n状態: 取得失敗 ({exc})\n")
+            continue
+
+        text = _text_from_bytes(url, data, ft, per_file_chars)
+        mime = _guess_mime(Path(httpx.URL(url).path or f"file.{ft}").name, ft)
+        name = Path(httpx.URL(url).path or f"file.{ft}").name
+        if session_id:
+            store_fetched_file(session_id, name, data, mime, text, url)
+        outputs.append(
+            f"=== 候補 {idx} ===\n"
+            f"URL: {url}\n"
+            f"タイトル: {title}\n"
+            f"形式: {mime}\n"
+            f"サイズ: {len(data)} bytes\n"
+            f"文字数: {len(text)}\n"
+            f"---\n{text}\n"
+        )
+
+    return "\n".join(outputs)
diff --git a/src/ollama_web/tools/pdf.py b/src/ollama_web/tools/pdf.py
new file mode 100644
index 0000000..8dc5abb
--- /dev/null
+++ b/src/ollama_web/tools/pdf.py
@@ -0,0 +1,88 @@
+"""PDF text extraction helpers."""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+from pypdf import PdfReader
+
+from ..config import settings
+
+
+def extract_pdf_text(data: bytes, max_chars: int | None = None) -> str:
+    """Extract text from PDF bytes.
+
+    Args:
+      data: Raw PDF file bytes.
+      max_chars: Maximum characters to return (default from settings).
+
+    Returns:
+      Extracted plain text. Errors are returned as short strings.
+    """
+    if not data:
+        return "Empty PDF data."
+
+    limit = max_chars if max_chars is not None and max_chars > 0 else settings.fetch_max_chars
+
+    try:
+        reader = PdfReader(stream=data)
+    except Exception as exc:  # noqa: BLE001
+        return f"Failed to parse PDF: {exc}"
+
+    parts: list[str] = []
+    for page in reader.pages:
+        try:
+            text = page.extract_text()
+            if text:
+                parts.append(text)
+        except Exception:  # noqa: BLE001
+            continue
+
+    full = "\n\n".join(parts)
+    if len(full) > limit:
+        full = full[:limit].rsplit(" ", 1)[0] + "\n…[truncated]"
+    return full or "No text could be extracted from the PDF."
+
+
+def pdf_to_text(url_or_path: str, max_chars: int | None = None) -> str:
+    """Extract text from a local PDF path or remote URL.
+
+    Args:
+      url_or_path: File path or URL pointing to a PDF.
+      max_chars: Maximum characters to return.
+
+    Returns:
+      Extracted plain text or an error message.
+    """
+    if not url_or_path or not url_or_path.strip():
+        return "Invalid PDF source."
+
+    src = url_or_path.strip()
+    data: bytes | None = None
+    if src.startswith("http://") or src.startswith("https://"):
+        import httpx
+        from ..config import settings as cfg  # local import to avoid top-level cycles
+
+        try:
+            with httpx.Client(
+                timeout=cfg.scrape_timeout,
+                follow_redirects=True,
+                headers={"User-Agent": "ollama-web/0.1 (+https://github.com/local)"},
+            ) as client:
+                resp = client.get(src)
+                resp.raise_for_status()
+                data = resp.content
+        except Exception as exc:  # noqa: BLE001
+            return f"Failed to fetch PDF: {exc}"
+    else:
+        path = Path(src)
+        if not path.exists():
+            return f"PDF not found: {src}"
+        try:
+            data = path.read_bytes()
+        except Exception as exc:  # noqa: BLE001
+            return f"Failed to read PDF: {exc}"
+
+    if data is None:
+        return "No PDF data loaded."
+    return extract_pdf_text(data, max_chars=max_chars)
diff --git a/src/ollama_web/tools/registry.py b/src/ollama_web/tools/registry.py
index 92b321c..2b6c7ae 100644
--- a/src/ollama_web/tools/registry.py
+++ b/src/ollama_web/tools/registry.py
@@ -6,6 +6,8 @@ import json
 from collections.abc import Callable, Sequence
 from dataclasses import dataclass, field
 
+from .fetch import search_and_fetch
+from .pdf import pdf_to_text
 from .scrape import scrape_url
 from .search import web_search
 
@@ -72,6 +74,8 @@ def default_registry() -> ToolRegistry:
     reg = ToolRegistry()
     reg.register(web_search)
     reg.register(scrape_url)
+    reg.register(search_and_fetch)
+    reg.register(pdf_to_text)
     return reg