app.py

"""

Level Bridge Chat -- Main app

Gradio Chat UI + FastAPI Bridge API endpoint, colocated.



Embedding (iframe):

  <iframe src="https://your-space.hf.space?campaign_name=X&industry=EC&cvr=2.1" ...></iframe>

"""

from __future__ import annotations
import os
import base64
from pathlib import Path

import gradio as gr
from fastapi import FastAPI, Request
from fastapi.responses import JSONResponse

from bridge_models import BridgeRequest, DashboardContext, Metrics
from bridge_service import process_request
from session_store import store

# ---------------------------------------------------------------------------
# FastAPI app
# ---------------------------------------------------------------------------
fastapi_app = FastAPI(title="Level Bridge Chat API")


@fastapi_app.post("/api/chat/bridge")
async def bridge_endpoint(request: Request):
    body = await request.json()
    try:
        req = BridgeRequest(**body)
    except Exception as e:
        return JSONResponse(
            status_code=422,
            content={"ok": False, "error_code": "VALIDATION_ERROR", "message": str(e)},
        )
    result = process_request(req)
    return JSONResponse(content=result.model_dump())


@fastapi_app.get("/healthz")
async def healthz():
    return {"ok": True, "sessions": store.count()}


# ---------------------------------------------------------------------------
# Helpers
# ---------------------------------------------------------------------------

def _parse_float(value: str | None) -> float | None:
    if value is None:
        return None
    try:
        return float(str(value).replace("%", "").replace("円", "").strip())
    except ValueError:
        return None


def _image_to_base64(image_path: str | None) -> str | None:
    if not image_path:
        return None
    try:
        with open(image_path, "rb") as f:
            data = f.read()
        ext = Path(image_path).suffix.lower().lstrip(".")
        mime = {"jpg": "jpeg", "jpeg": "jpeg", "png": "png", "webp": "webp", "gif": "gif"}.get(ext, "png")
        return f"data:image/{mime};base64," + base64.b64encode(data).decode("utf-8")
    except Exception:
        return None


def _format_response(result) -> str:
    """Format BridgeResponse or BridgeErrorResponse as markdown chat message."""
    if not result.ok:
        msg = f"**エラー**: {result.message}"
        if hasattr(result, "fallback") and result.fallback and result.fallback.get("next_level_preview"):
            nlp = result.fallback["next_level_preview"]
            if nlp.get("needed_info"):
                items = ", ".join(i["label"] for i in nlp["needed_info"])
                msg += f"\n\n次に **{items}** を追加すると提案が可能になります。"
        return msg

    b = result.best_now
    n = result.next_level_preview
    level_label = {"level1": "Lv.1（基本情報）", "level2": "Lv.2（数値あり）", "level3": "Lv.3（画像あり）"}.get(
        result.inferred_level, result.inferred_level
    )
    confidence_label = {"low": "低", "mid": "中", "high": "高"}.get(b.confidence, b.confidence)

    lines = [
        f"### 現在の提案 [{level_label} / 確信度: {confidence_label}]",
        "",
        b.summary,
        "",
        "**推奨アクション**",
    ]
    for i, action in enumerate(b.actions, 1):
        lines.append(f"{i}. {action}")

    if n.next_level:
        lines += [
            "",
            "---",
            f"### 次レベル（{n.next_level}）で可能になること",
            "",
        ]
        if n.needed_info:
            items = "、".join(f"`{i.label}`（例: {i.example}）" for i in n.needed_info)
            lines.append(f"**必要な情報**: {items}")
            lines.append("")
        for item in n.what_will_be_possible:
            lines.append(f"- {item}")
        if n.expected_impact:
            lines += ["", f"*{n.expected_impact}*"]
    else:
        lines += ["", "---", "*全情報が揃っています。現在の提案は最高精度です。*"]

    if result.follow_up_question:
        lines += ["", f"**{result.follow_up_question}**"]

    return "\n".join(lines)


# ---------------------------------------------------------------------------
# Gradio UI
# ---------------------------------------------------------------------------

def build_gradio_ui() -> gr.Blocks:
    with gr.Blocks(title="Level Bridge Chat") as demo:
        gr.Markdown("## Level Bridge Chat")
        gr.Markdown(
            "ダッシュボードの情報をもとに、広告改善提案を行います。"
            " 情報を追加するたびに提案精度が向上します。"
        )

        # State
        session_id_state = gr.State(None)
        context_state = gr.State({})

        # Level indicator
        level_display = gr.Markdown("**情報レベル**: 初期化中...")

        chatbot = gr.Chatbot(
            label="提案チャット",
            height=480,
        )

        with gr.Row():
            msg_input = gr.Textbox(
                placeholder="メッセージを入力（例: CVR 2.1%、CTR 0.8% です）",
                label="メッセージ",
                scale=4,
                show_label=False,
            )
            send_btn = gr.Button("送信", variant="primary", scale=1)

        with gr.Row():
            image_upload = gr.Image(
                label="クリエイティブ画像（任意）",
                type="filepath",
                height=160,
            )

        # --- Core chat function ---
        def send_message(

            message: str,

            history: list,

            session_id: str | None,

            image_path: str | None,

            ctx: dict,

        ):
            if not message.strip() and not image_path:
                return history, session_id, "", None, ctx, gr.update()

            image_b64 = _image_to_base64(image_path)

            # Build dashboard_context only when there are new values
            dc = None
            if ctx or image_b64:
                metrics_obj = None
                if ctx.get("cvr") or ctx.get("ctr") or ctx.get("cpa"):
                    metrics_obj = Metrics(
                        cvr=ctx.get("cvr"),
                        ctr=ctx.get("ctr"),
                        cpa=ctx.get("cpa"),
                    )
                dc = DashboardContext(
                    campaign_name=ctx.get("campaign_name"),
                    industry=ctx.get("industry"),
                    metrics=metrics_obj,
                    image_base64=image_b64,
                )

            req = BridgeRequest(
                session_id=session_id,
                message=message,
                dashboard_context=dc,
            )
            result = process_request(req)
            response_text = _format_response(result)

            new_session_id = getattr(result, "session_id", session_id)
            new_level = getattr(result, "inferred_level", "level1")
            level_label = {
                "level1": "Lv.1（基本情報のみ）",
                "level2": "Lv.2（定量データあり）",
                "level3": "Lv.3（画像あり・最高精度）",
            }.get(new_level, new_level)

            history = history + [
                {"role": "user", "content": message or "（画像を送信）"},
                {"role": "assistant", "content": response_text},
            ]

            # Clear context after first send (already stored in session)
            new_ctx = {}

            return (
                history,
                new_session_id,
                "",  # clear message input
                None,  # clear image
                new_ctx,
                gr.update(value=f"**情報レベル**: {level_label}"),
            )

        # --- Auto-initialize from URL query params ---
        def on_load(request: gr.Request):
            params = dict(request.query_params)
            ctx = {}
            if params.get("campaign_name"):
                ctx["campaign_name"] = params["campaign_name"]
            if params.get("industry"):
                ctx["industry"] = params["industry"]
            if params.get("cvr"):
                ctx["cvr"] = _parse_float(params["cvr"])
            if params.get("ctr"):
                ctx["ctr"] = _parse_float(params["ctr"])
            if params.get("cpa"):
                ctx["cpa"] = _parse_float(params["cpa"])

            if not ctx:
                return [], None, ctx, gr.update(value="**情報レベル**: 未初期化（URLパラメータなし）")

            # Auto-send Turn 1 with dashboard context
            metrics_obj = None
            if ctx.get("cvr") or ctx.get("ctr") or ctx.get("cpa"):
                metrics_obj = Metrics(cvr=ctx.get("cvr"), ctr=ctx.get("ctr"), cpa=ctx.get("cpa"))

            dc = DashboardContext(
                campaign_name=ctx.get("campaign_name"),
                industry=ctx.get("industry"),
                metrics=metrics_obj,
            )
            req = BridgeRequest(session_id=None, message="", dashboard_context=dc)
            result = process_request(req)
            response_text = _format_response(result)

            new_session_id = getattr(result, "session_id", None)
            new_level = getattr(result, "inferred_level", "level1")
            level_label = {
                "level1": "Lv.1（基本情報のみ）",
                "level2": "Lv.2（定量データあり）",
                "level3": "Lv.3（画像あり・最高精度）",
            }.get(new_level, new_level)

            initial_history = [{"role": "assistant", "content": response_text}]

            return (
                initial_history,
                new_session_id,
                {},
                gr.update(value=f"**情報レベル**: {level_label}"),
            )

        # Wire events
        demo.load(
            on_load,
            inputs=None,
            outputs=[chatbot, session_id_state, context_state, level_display],
        )

        send_btn.click(
            send_message,
            inputs=[msg_input, chatbot, session_id_state, image_upload, context_state],
            outputs=[chatbot, session_id_state, msg_input, image_upload, context_state, level_display],
        )

        msg_input.submit(
            send_message,
            inputs=[msg_input, chatbot, session_id_state, image_upload, context_state],
            outputs=[chatbot, session_id_state, msg_input, image_upload, context_state, level_display],
        )

    return demo


# ---------------------------------------------------------------------------
# Mount and launch
# ---------------------------------------------------------------------------
gradio_ui = build_gradio_ui()
app = gr.mount_gradio_app(fastapi_app, gradio_ui, path="/")


if __name__ == "__main__":
    import uvicorn
    port = int(os.environ.get("PORT", "7860"))
    uvicorn.run(app, host="0.0.0.0", port=port, log_level="info")