Blame for routes/chat.py

import json
from fastapi import APIRouter, HTTPException, Depends
from fastapi.responses import StreamingResponse
from pydantic import BaseModel
from firebase_admin import firestore
from utils.rate_limiter import with_rate_limit
from utils.groq_client import call_groq_stream, LIGHT_MODEL, HEAVY_MODEL
from utils.pinecone_client import query_book_context
from utils import business_type

router = APIRouter()


# ─── Schemas ──────────────────────────────────────────────────────────────────

class Message(BaseModel):
    role: str    # "user" or "assistant"
    content: str


class ChatRequest(BaseModel):
    businessId: str
    sessionId: str
    message: str
    history: list[Message] = []
    useHeavyModel: bool = False   # True for AI Consultation sessions


# ─── Helpers ──────────────────────────────────────────────────────────────────

HISTORY_LIMIT = 10          # last N messages kept for context window efficiency
HISTORY_CONTENT_CHARS = 4000  # per-message cap so one huge paste can't bloat every call


def build_messages(system_prompt: str, history: list[Message], message: str) -> list[dict]:
    """
    System prompt + recent history + the new message. Only 'user' and
    'assistant' roles are accepted from the client: history comes from the
    browser, so a 'system' role in it would let a caller inject instructions.
    """
    messages = [{"role": "system", "content": system_prompt}]
    for msg in history[-HISTORY_LIMIT:]:
        if msg.role in ("user", "assistant"):
            messages.append({"role": msg.role, "content": msg.content[:HISTORY_CONTENT_CHARS]})
    messages.append({"role": "user", "content": message})
    return messages

def _get_business_context(user_id: str, business_id: str) -> str:
    """Builds a system context string from BFF + profile_output in Firestore."""
    db = firestore.client()
    biz_ref = (
        db.collection("user_data")
        .document(user_id)
        .collection("businesses")
        .document(business_id)
    )

    # Fetch BFF section 1 (WHAT) for quick context
    s1 = biz_ref.collection("bff").document("section1").get()
    # Fetch business identity if generated
    identity = biz_ref.collection("profile_output").document("businessIdentity").get()

    ctx_parts = []

    if s1.exists:
        d = s1.to_dict()
        t = business_type.describe(d)
        ctx_parts.append(
            f"Business: {d.get('businessName', 'Unknown')}, "
            f"Sector: {t['sector']}, "
            f"Type of business: {t['subsector']}, "
            f"Niche: {t['niche']}"
        )

    if identity.exists:
        d = identity.to_dict()
        ctx_parts.append(f"Summary: {d.get('summary', '')}")
        ctx_parts.append(f"Value Proposition: {d.get('valueProposition', '')}")

    return "\n".join(ctx_parts) if ctx_parts else "Business profile not yet generated."


# ─── Route: Chat message (streaming SSE) ─────────────────────────────────────

@router.post("/message")
def chat_message(
    body: ChatRequest,
    user: dict = Depends(with_rate_limit("chat")),
):
    """
    Accepts a user message and returns a streaming SSE response.
    For passive AI Chat: uses light model (LIGHT_MODEL).
    For AI Consultation: uses heavy model (HEAVY_MODEL).

    Stream format (SSE):
      data: {"type": "thinking", "token": "..."}   ← reasoning (heavy model only)
      data: {"type": "response", "token": "..."}   ← main response
      data: {"type": "done"}                        ← stream end
    """
    user_id = user.get("uid")
    model   = HEAVY_MODEL if body.useHeavyModel else LIGHT_MODEL

    business_context = _get_business_context(user_id, body.businessId)

    # Retrieve relevant passages from "Built on Rock" based on user's message
    book_ctx = query_book_context(body.message, top_k=2)

    system_prompt = f"""You are the VenThrive Business Assistant — a business consultant trained in the Business Footing Framework (BFF) from "Built on Rock" by Iyanuoluwa Joseph Akinnusi. You help Nigerian entrepreneurs build successful businesses using the WHAT-WHY-HOW philosophy.

BUSINESS CONTEXT:
{business_context}
{f"{chr(10)}RELEVANT BFF GUIDANCE FROM 'BUILT ON ROCK':{chr(10)}{book_ctx}" if book_ctx else ""}

RULES:
1. STRICTLY BUSINESS: Stay focused on the entrepreneur's business. If asked personal questions, redirect warmly.
2. NIGERIAN CONTEXT: Reference real Nigerian examples — banks (GTBank, First Bank, Moniepoint), platforms (Jumia, Konga, Flutterwave), markets (Balogun, Alaba), cities and states.
3. BFF PHILOSOPHY: When relevant, ground advice in the WHAT-WHY-HOW framework from "Built on Rock".
4. PLAIN ENGLISH: No business jargon without explanation. Write like you're talking to a smart friend.
5. CONCISE: 5-6 sentences max per response. Use bullet points for lists.
5. ACTIONABLE: Every response should tell the entrepreneur what to DO next, not just what things mean.
"""

    messages = build_messages(system_prompt, body.history, body.message)

    def stream():
        try:
            # call_groq_stream handles key rotation on 429
            stream_response = call_groq_stream(
                system_prompt=system_prompt,
                user_prompt=body.message,
                model=model,
                max_tokens=512,
                fixture="chat_response",
                messages=messages,
            )
            for chunk in stream_response:
                token = chunk.choices[0].delta.content
                if token:
                    yield f"data: {json.dumps({'type': 'response', 'token': token})}\n\n"

            yield f"data: {json.dumps({'type': 'done'})}\n\n"

        except Exception as e:
            yield f"data: {json.dumps({'type': 'error', 'message': 'Something went wrong. Please try again.'})}\n\n"

    return StreamingResponse(stream(), media_type="text/event-stream")