backend/main.py

"""FastAPI application wiring orchestration between planner, tools, and UI."""

from __future__ import annotations
import csv
import json
import logging
import os
import threading
import time
import warnings
from functools import lru_cache
from pathlib import Path
from typing import Any, Dict, List, Optional, Tuple

from fastapi import FastAPI, HTTPException, Query
from fastapi.responses import FileResponse, RedirectResponse
from fastapi.staticfiles import StaticFiles
from pydantic import BaseModel

try:
    from huggingface_hub import CommitScheduler, hf_hub_download
except ImportError:  # pragma: no cover - optional dependency
    CommitScheduler = None  # type: ignore
    hf_hub_download = None  # type: ignore

from .data import load_default_catalog
from .tools import (
    AnalysisEngine,
    FacultyByTopicBlueprint,
    LocationBlueprint,
    CenterBlueprint,
    AdvisorshipBlueprint,
    PersonLookupBlueprint,
    StaffSupportBlueprint,
    UpcomingEventsBlueprint,
    OfficeHoursBlueprint,
    BlueprintResult,
)
from .responders import LLMResponder, Responder
from .providers import (
    BaseLLM,
    ProviderConfig,
    available_providers,
    get_provider,
    normalize_provider_name,
)
from .data.utils import canonicalize_name
from .mcp import (
    Action,
    PlannerContext,
)
from .mcp.tool_executor import ToolExecutor
from .mcp.vanilla_planner import VanillaLMPlanner
from .mcp.vanilla_responder import VanillaLMResponder
from .msai_bridge import warm_load
from .mcp.context_resolver import (
    is_affirmation,
    resolve as resolve_context,
    strip_context_on_topic_switch,
)

BASE_DIR = Path(__file__).resolve().parent
ARCHIVE_DIR = BASE_DIR.parent / "Archive"
DATA_DIR = BASE_DIR / "storage"
DATA_DIR.mkdir(parents=True, exist_ok=True)
FRONTEND_DIST = BASE_DIR.parent / "frontend" / "dist"
FRONTEND_INDEX = FRONTEND_DIST / "index.html"

HISTORY_FILE = DATA_DIR / "chat_history.jsonl"
USAGE_FILE = DATA_DIR / "usage_metrics.jsonl"

DEFAULT_SESSION = "default"

app = FastAPI(title="Northwestern CS Kiosk (Vanilla LM)")

_orchestrator_lock = threading.Lock()
logger = logging.getLogger(__name__)
_usage_metrics_scheduler = None
_entity_names: List[str] = []


class QueryPayload(BaseModel):
    question: str
    session_id: Optional[str] = None
    provider: Optional[str] = None


PROVIDER_ENV_SETTINGS: Dict[str, Dict[str, Optional[str]]] = {
    "claude": {
        "api_key": "ANTHROPIC_API_KEY",
        "model": "ANTHROPIC_MODEL",
        "base_url": "ANTHROPIC_BASE_URL",
        "default_model": "claude-haiku-4-5",
    },
    "gpt": {
        "api_key": "OPENAI_API_KEY",
        "model": "OPENAI_MODEL",
        "base_url": "OPENAI_BASE_URL",
        "default_model": "gpt-4.1-mini",
    },
    "gemini": {
        "api_key": "GEMINI_API_KEY",
        "model": "GEMINI_MODEL",
        "base_url": "GEMINI_BASE_URL",
        "default_model": "gemini-2.0-flash",
    },
    "echo": {
        "api_key": None,
        "model": None,
        "base_url": None,
        "default_model": "echo",
    },
}

def _load_entity_names() -> None:
    """Scrape entity names from Archive folder at startup and store in memory."""
    global _entity_names
    
    def _extract_names_from_csv(filepath: Path) -> List[str]:
        """Extract names from CSV file based on detected column headers."""
        names = []
        try:
            with open(filepath, 'r', encoding='utf-8') as f:
                reader = csv.DictReader(f)
                if reader.fieldnames is None:
                    return names
                
                fieldnames = reader.fieldnames
                name_columns = []
                
                # Check for various name column patterns
                for field in fieldnames:
                    field_lower = field.lower()
                    if field_lower == 'name' or field_lower == 'assignee name':
                        name_columns = [field]
                        break
                    elif field_lower == 'first name':
                        name_columns.append(field)
                    elif field_lower == 'last name':
                        name_columns.insert(0, field)
                
                # Extract names
                for row in reader:
                    if name_columns:
                        # Single name column
                        if len(name_columns) == 1 and row.get(name_columns[0]):
                            name = row[name_columns[0]].strip()
                            if name and name.upper() != 'NA':
                                names.append(name)
                        # First name and last name columns
                        elif len(name_columns) == 2:
                            last_name = row.get(name_columns[0], '').strip()
                            first_name = row.get(name_columns[1], '').strip()
                            if (last_name or first_name) and last_name.upper() != 'NA' and first_name.upper() != 'NA':
                                if last_name and first_name:
                                    full_name = f"{first_name} {last_name}"
                                elif first_name:
                                    full_name = first_name
                                else:
                                    full_name = last_name
                                if full_name:
                                    names.append(full_name)
        except Exception as e:
            logger.warning(f"Error reading CSV {filepath}: {e}")
        return names
    
    def _extract_names_from_text(filepath: Path) -> List[str]:
        """Extract names from text files with 'Name: ' prefix."""
        names = []
        try:
            with open(filepath, 'r', encoding='utf-8') as f:
                for line in f:
                    line = line.strip()
                    if line.startswith('Name:'):
                        name = line.replace('Name:', '').strip()
                        if name:
                            names.append(name)
        except Exception as e:
            logger.warning(f"Error reading text file {filepath}: {e}")
        return names
    
    try:
        archive_dir = ARCHIVE_DIR
        if not archive_dir.exists():
            logger.warning(f"Archive directory not found at {archive_dir}")
            _entity_names = []
            return
        
        all_names: set = set()
        file_count = 0
        
        # Process all files in Archive
        for filepath in sorted(archive_dir.iterdir()):
            if filepath.is_file():
                if filepath.suffix.lower() == '.csv':
                    names = _extract_names_from_csv(filepath)
                    all_names.update(names)
                    file_count += 1
                    logger.debug(f"Extracted {len(names)} names from {filepath.name}")
                
                elif filepath.suffix.lower() == '.txt':
                    names = _extract_names_from_text(filepath)
                    all_names.update(names)
                    file_count += 1
                    logger.debug(f"Extracted {len(names)} names from {filepath.name}")
        
        # Convert to sorted list, removing duplicates
        _entity_names = sorted(all_names)
        logger.info(f"Scraped {len(_entity_names)} unique entity names from {file_count} files in Archive")
    
    except Exception as e:
        logger.error(f"Failed to scrape entity names from Archive: {e}")
        _entity_names = []


def _load_env_once() -> None:
    """Load environment variables from .env exactly once."""

    if getattr(_load_env_once, "_loaded", False):
        return

    env_path = os.getenv("KIOSK_ENV_FILE")
    if not env_path:
        default_path = BASE_DIR / ".env"
        env_path = str(default_path) if default_path.exists() else ".env"

    try:
        from dotenv import load_dotenv  # type: ignore
    except ImportError:
        _load_env_once._loaded = True
        return

    load_dotenv(env_path, override=False)
    _load_env_once._loaded = True

def _maybe_download_existing_metrics() -> None:
    """Download existing usage metrics from HF dataset on startup."""
    repo_id = os.getenv("KIOSK_HF_DATASET_REPO", "").strip()
    if not repo_id or hf_hub_download is None:
        return
    _load_env_once()
    token = _get_env_value("KIOSK_HF_TOKEN") or os.getenv("HF_TOKEN", "").strip()
    path_in_repo = os.getenv("KIOSK_HF_DATASET_PATH", "chat_history").strip()
    filename = f"{path_in_repo}/{USAGE_FILE.name}" if path_in_repo else USAGE_FILE.name
    try:
        import shutil
        downloaded = hf_hub_download(
            repo_id=repo_id, repo_type="dataset", filename=filename, token=token or None,
        )
        USAGE_FILE.parent.mkdir(parents=True, exist_ok=True)
        shutil.copy(downloaded, USAGE_FILE)
        logger.info("Downloaded usage metrics from HF: repo=%s file=%s", repo_id, filename)
    except Exception as exc:
        logger.info("No existing metrics to download (starting fresh): %s", exc)


def _maybe_download_existing_history() -> None:
    """Download existing chat history from HF dataset on startup."""
    repo_id = os.getenv("KIOSK_HF_DATASET_REPO", "").strip()
    if not repo_id or hf_hub_download is None:
        return
    _load_env_once()
    token = _get_env_value("KIOSK_HF_TOKEN") or os.getenv("HF_TOKEN", "").strip()
    path_in_repo = os.getenv("KIOSK_HF_DATASET_PATH", "chat_history").strip()
    filename = f"{path_in_repo}/{HISTORY_FILE.name}" if path_in_repo else HISTORY_FILE.name
    try:
        import shutil
        downloaded = hf_hub_download(
            repo_id=repo_id, repo_type="dataset", filename=filename, token=token or None,
        )
        HISTORY_FILE.parent.mkdir(parents=True, exist_ok=True)
        shutil.copy(downloaded, HISTORY_FILE)
        logger.info("Downloaded chat history from HF: repo=%s file=%s", repo_id, filename)
    except Exception as exc:
        logger.info("No existing chat history to download (starting fresh): %s", exc)


def _maybe_start_usage_metrics_sync() -> None:
    """Start optional HF dataset syncing for chat history and usage metrics."""
    global _usage_metrics_scheduler
    if _usage_metrics_scheduler is not None:
        return
    repo_id = os.getenv("KIOSK_HF_DATASET_REPO", "").strip()
    if not repo_id:
        return
    if CommitScheduler is None:
        warnings.warn("huggingface_hub not installed; skipping HF sync.")
        return
    _load_env_once()
    token = _get_env_value("KIOSK_HF_TOKEN") or os.getenv("HF_TOKEN", "").strip()
    path_in_repo = os.getenv("KIOSK_HF_DATASET_PATH", "chat_history").strip()
    interval_minutes = float(os.getenv("KIOSK_HF_SYNC_INTERVAL_MINUTES", "10"))
    try:
        _usage_metrics_scheduler = CommitScheduler(
            repo_id=repo_id,
            repo_type="dataset",
            folder_path=str(DATA_DIR),
            path_in_repo=path_in_repo,
            token=token or None,
            allow_patterns=[HISTORY_FILE.name, USAGE_FILE.name],
            every=interval_minutes,
        )
        logger.info(
            "Started HF CommitScheduler for chat_history and usage_metrics: repo=%s path=%s interval=%s",
            repo_id, path_in_repo or ".", interval_minutes,
        )
    except Exception as exc:  # pragma: no cover - defensive
        warnings.warn(f"Unable to start HF sync: {exc}")


def _run_startup_tasks_in_background() -> None:
    """Run HF download and sync in a background thread so the server starts immediately."""
    def _run() -> None:
        try:
            _maybe_download_existing_metrics()
            _maybe_download_existing_history()
            _maybe_start_usage_metrics_sync()
        except Exception as exc:  # pragma: no cover
            logger.warning("Background startup tasks failed: %s", exc)

    t = threading.Thread(target=_run, daemon=True)
    t.start()


def _get_env_value(name: Optional[str]) -> str:
    """
    Read environment variables with an HF Spaces secret fallback.

    HF Secrets expose values as HF_<NAME>, so check both keys.
    """
    if not name:
        return ""
    direct = os.getenv(name, "").strip()
    if direct:
        return direct
    return os.getenv(f"HF_{name}", "").strip()


def _is_placeholder(value: Optional[str]) -> bool:
    if not value:
        return True
    lowered = value.strip().lower()
    return lowered.startswith("your-") or lowered in {"changeme", "placeholder"}


_load_entity_names()
_run_startup_tasks_in_background()


def _build_client_from_env(provider: str, model_override: Optional[str]) -> Optional[BaseLLM]:
    canonical = normalize_provider_name(provider)
    settings = PROVIDER_ENV_SETTINGS.get(canonical)
    if not settings:
        warnings.warn(f"Provider '{provider}' not recognized; falling back to echo responder.")
        return None

    timeout = int(os.getenv("KIOSK_LLM_TIMEOUT", "60"))
    max_tokens_raw = os.getenv("KIOSK_LLM_MAX_TOKENS", "").strip()
    max_tokens = int(max_tokens_raw) if max_tokens_raw.isdigit() else None
    api_env = settings.get("api_key")
    model_env = settings.get("model")
    base_url_env = settings.get("base_url")
    default_model = settings.get("default_model") or ""

    if api_env:
        api_key = _get_env_value(api_env)
        if not api_key or _is_placeholder(api_key):
            warnings.warn(f"{api_env} not set; falling back to echo responder.")
            return None
    else:
        api_key = "local-echo"
    model = model_override or (_get_env_value(model_env) if model_env else "") or default_model
    base_url = _get_env_value(base_url_env) if base_url_env else ""

    config = ProviderConfig(
        api_key=api_key,
        model=model,
        timeout_sec=timeout,
        base_url=base_url or None,
        max_tokens=max_tokens,
    )

    try:
        return get_provider(canonical, config)
    except ValueError as exc:
        warnings.warn(str(exc))
        return None


def _build_responder(
    provider: Optional[str],
    model_override: Optional[str],
) -> LLMResponder:
    _load_env_once()
    system_prompt = os.getenv(
        "KIOSK_LLM_SYSTEM_PROMPT",
        "You are a conversational receptionist for the Northwestern CS Kiosk whose responses are spoken aloud. Speak naturally and never include stage directions or annotations.",
    )
    style = os.getenv("KIOSK_LLM_STYLE", "Be very brief. One or two sentences max. No long lists—summarize top 2-3 items only.")

    provider_name = provider or os.getenv("KIOSK_LLM_PROVIDER", "anthropic")
    model_override = model_override if provider else (model_override or os.getenv("KIOSK_LLM_MODEL"))

    client = _build_client_from_env(provider_name, model_override)
    canonical = normalize_provider_name(provider_name)

    if client:
        return LLMResponder(
            client=client,
            system_prompt=system_prompt,
            style_guidelines=style,
            provider_id=canonical,
        )
    warnings.warn("LLM provider not configured; using echo responder for kiosk responses.")
    return LLMResponder(
        system_prompt=system_prompt,
        style_guidelines=style,
        provider_id="echo",
    )


def _default_responder_from_env() -> Responder:
    try:
        return _build_responder(None, None)
    except RuntimeError as exc:
        warnings.warn(f"Failed to initialize LLM responder: {exc}")
        return LLMResponder(provider_id="echo")


def _create_planner() -> LLMActionPlanner:
    provider = os.getenv("KIOSK_PLANNER_PROVIDER") or os.getenv("KIOSK_LLM_PROVIDER", "anthropic")
    model_override = os.getenv("KIOSK_PLANNER_MODEL") or os.getenv("KIOSK_LLM_MODEL")
    client = _build_client_from_env(provider, model_override)
    if not client:
        raise RuntimeError("LLM planner requires a configured provider (set KIOSK_LLM_PROVIDER/KEY).")
    schemas = get_all_tool_schemas()
    return LLMActionPlanner(client, schemas=schemas, entity_names=_entity_names)


class ConversationOrchestrator:
    """Glue class that ties planner, executor, and responder together."""

    def __init__(self, engine: AnalysisEngine, responder: Optional[Responder] = None) -> None:
        _load_env_once()
        self.engine = engine
        self.planner = VanillaLMPlanner(available_names=_entity_names)
        self.responder = responder or VanillaLMResponder(self.planner)
        self.executor = ToolExecutor(engine)
        self.last_subject: Optional[str] = None
        self._faculty_lookup = self._build_name_lookup("faculty")
        self._student_lookup = self._build_name_lookup("students")
        self.provider_id = getattr(self.responder, "provider_id", "vanilla")

    def answer(
        self,
        question: str,
        context: Optional[PlannerContext] = None,
        resolved_input: Optional[Any] = None,
    ) -> Tuple[str, BlueprintResult, Action]:
        if context is None:
            context = PlannerContext(last_subject=self.last_subject)

        # Affirmation short-circuit: "yes please" when we offered to look up location
        if is_affirmation(question) and context.last_subject:
            last_answer = ""
            if context.short_history:
                last_answer = (context.short_history[-1].get("answer") or "").lower()
            if any(
                x in last_answer
                for x in ("would you like", "look up", "find", "room number", "office")
            ):
                self.planner.lm_routing_succeeded = False
                actions = [Action("lookup_location", {"use_last_subject": True})]
            else:
                actions = self.planner.plan(question, context)
        else:
            actions = self.planner.plan(question, context)

        if not actions:
            actions = [Action("noop", {"message": "I'm not sure how to help with that yet."})]

        # Inject resolved day (e.g. "F" → "friday", "today" → "wednesday") when planner returns lookup_office_hours without day
        if actions and resolved_input and getattr(resolved_input, "resolved_day", None):
            for act in actions:
                if act.type == "lookup_office_hours" and not act.arguments.get("day"):
                    act.arguments["day"] = resolved_input.resolved_day

        # If the planner returned multiple actions, execute them all and merge
        # their blueprint outputs so the final responder call can reason over
        # the combined facts. If only one action was returned, fall back to
        # the prior single-action flow (including the name-based multi-lookup).
        if len(actions) > 1:
            merged_facts: List = []
            merged_notes: List[str] = []
            ran: List[str] = []
            for act in actions:
                ran.append(act.type)
                sub_result = self.executor.execute(act, context)
                merged_facts.extend(sub_result.facts)
                for note in sub_result.notes:
                    if note not in merged_notes:
                        merged_notes.append(note)
            result = BlueprintResult("composite", {}, facts=merged_facts, notes=merged_notes)
            action = Action("composite", {"actions": [a.to_dict() for a in actions], "merged_actions": ran})
        else:
            action = actions[0]
            # If the planner provided a person name, run multiple person-oriented
            # lookup tools (faculty and student lookups) and merge their results so
            # the responder can provide a combined answer rather than relying on a
            # single-tool decision from the planner.
            name_like = None
            if isinstance(action.arguments, dict):
                for key in ("name", "person", "student", "faculty"):
                    val = action.arguments.get(key)
                    if val:
                        name_like = val
                        break

            if name_like:
                # Only execute the single action the planner suggested.
                result = self.executor.execute(action, context)

                # If the planner's chosen tool returned nothing, attempt a
                # conservative disambiguation using the local catalog: prefer a
                # single explicit faculty/student resolution and re-run the
                # planner's original action targeted at that resolved entity.
                if not result.facts:
                    canonical = canonicalize_name(name_like)
                    faculty_match = self._faculty_lookup.get(canonical)
                    student_match = self._student_lookup.get(canonical)
                    if faculty_match and not student_match:
                        action.arguments.pop("name", None)
                        action.arguments.pop("person", None)
                        action.arguments["faculty"] = faculty_match
                        result = self.executor.execute(action, context)
                    elif student_match and not faculty_match:
                        action.arguments.pop("name", None)
                        action.arguments.pop("person", None)
                        action.arguments["student"] = student_match
                        result = self.executor.execute(action, context)
                    elif faculty_match and student_match:
                        action = Action(
                            "noop",
                            {
                                "message": (
                                    f"I found both a faculty member and a student named {name_like}. "
                                    "Do you mean the faculty member or the student?"
                                )
                            },
                        )
                        result = BlueprintResult("noop", action.arguments, facts=[], notes=[action.arguments.get("message")])
            else:
                result = self.executor.execute(action, context)
        response_text = self.responder.render(question, result.name, result)

        subject = self._select_subject_from_result(result)
        if subject:
            self.last_subject = subject
        else:
            for key in ("name", "student", "faculty"):
                if key in action.arguments and action.arguments[key]:
                    self.last_subject = action.arguments[key]
                    break
        return response_text, result, action

    def ensure_responder(self, provider: Optional[str], model_override: Optional[str] = None) -> None:
        del model_override
        if provider and normalize_provider_name(provider) not in (None, "vanilla"):
            warnings.warn(f"Ignoring provider '{provider}'; kiosk_vanilla uses local vanilla LM only.")
        self.provider_id = "vanilla"

    @staticmethod
    def _infer_subject(result: BlueprintResult) -> Optional[str]:
        if not result.facts:
            return None
        return result.facts[0].subject

    def _build_name_lookup(self, entity_name: str) -> Dict[str, str]:
        mapping: Dict[str, str] = {}
        entity = self.engine.catalog.try_get(entity_name)
        if not entity:
            return mapping
        for row in entity.records:
            name = row.get("Name")
            if not name:
                continue
            mapping[canonicalize_name(name)] = name
        return mapping

    def _select_subject_from_result(self, result: BlueprintResult) -> Optional[str]:
        candidates: List[str] = []
        for fact in result.facts:
            if isinstance(fact.subject, str):
                candidates.append(fact.subject)
            if isinstance(fact.value, str):
                candidates.append(fact.value)
        for candidate in candidates:
            canonical = canonicalize_name(candidate)
            if canonical in self._faculty_lookup:
                return self._faculty_lookup[canonical]
        for candidate in candidates:
            canonical = canonicalize_name(candidate)
            if canonical in self._student_lookup:
                return self._student_lookup[canonical]
        return self._infer_subject(result)


def _append_json_line(path: Path, payload: Dict[str, Any]) -> None:
    path.parent.mkdir(parents=True, exist_ok=True)
    with path.open("a", encoding="utf-8") as handle:
        handle.write(json.dumps(payload, ensure_ascii=False) + "\n")


def record_history(
    *,
    session_id: str,
    question: str,
    answer: str,
    blueprint: str,
    metadata: Dict[str, Any],
    facts: List[Dict[str, Any]],
    notes: List[str],
    action: Dict[str, Any],
) -> float:
    timestamp = time.time()
    payload = {
        "timestamp": timestamp,
        "session_id": session_id,
        "question": question,
        "answer": answer,
        "blueprint": blueprint,
        "facts": facts,
        "notes": notes,
        "usage": metadata,
        "action": action,
    }
    _append_json_line(HISTORY_FILE, payload)
    usage_entry = {
        "timestamp": payload["timestamp"],
        "session_id": session_id,
        "blueprint": blueprint,
        "question": question,
    }
    usage_entry.setdefault("action_type", action.get("type"))
    _append_json_line(USAGE_FILE, usage_entry)
    logger.info(
        "Recorded usage metrics for session=%s blueprint=%s question=%s",
        session_id,
        blueprint,
        question,
    )
    return timestamp


def load_history(session_id: str) -> List[Dict[str, Any]]:
    if not HISTORY_FILE.exists():
        return []
    rows: List[Dict[str, Any]] = []
    with HISTORY_FILE.open(encoding="utf-8") as handle:
        for line in handle:
            try:
                record = json.loads(line)
            except json.JSONDecodeError:
                continue
            if record.get("session_id") == session_id:
                rows.append(record)
    rows.sort(key=lambda row: row.get("timestamp", 0))
    return rows


def build_planner_context_from_history(history: List[Dict[str, Any]]) -> PlannerContext:
    """Build session context: full history, short window, and topic-aware follow-up state."""
    if not history:
        return PlannerContext()

    # Full history: last ~20 turns to bound tokens (question + answer per turn)
    cap = 20
    full_history: List[Dict[str, Any]] = []
    for rec in history[-cap:]:
        full_history.append({
            "question": rec.get("question", ""),
            "answer": rec.get("answer", ""),
        })

    # Short history: last 3–4 turns with action for planner
    short_history: List[Dict[str, Any]] = []
    for rec in history[-4:]:
        short_history.append({
            "question": rec.get("question", ""),
            "answer": rec.get("answer", ""),
            "action": rec.get("action"),
        })
    short_history = short_history[-3:]

    last = history[-1]
    action = last.get("action") or {}
    args = action.get("arguments") or {}
    facts = last.get("facts") or []
    action_type = (action.get("type") or "").lower()

    # Infer topic and subject from last turn
    topic: Optional[str] = None
    subject: Optional[str] = None
    last_class: Optional[str] = None

    # Subject from facts (canonical) or action args
    if facts and isinstance(facts[0], dict) and facts[0].get("subject"):
        subject = facts[0]["subject"]
    if not subject:
        for key in ("name", "person", "student", "faculty", "class_name", "course"):
            if args.get(key):
                subject = args[key]
                break

    # Topic and last_class from action type
    if action_type == "lookup_office_hours":
        topic = "office_hours"
        cls_val = args.get("class_name") or args.get("course") or ""
        if cls_val:
            last_class = cls_val
        elif subject and (any(c.isdigit() for c in subject) or subject.upper().startswith("CS")):
            last_class = subject
    elif action_type in (
        "lookup_person",
        "lookup_location",
        "lookup_center",
        "lookup_advisorship",
        "lookup_faculty_topic",
    ):
        # Heuristic: advisorship with student arg → student; else professor
        if action_type == "lookup_advisorship" and args.get("student"):
            topic = "student"
        elif action_type == "lookup_faculty_topic" or action_type == "lookup_center":
            topic = "professor"
        else:
            topic = "professor"  # person/location/advisorship default to professor

    # last_subject for executor (use_last_subject)
    last_subject = subject

    return PlannerContext(
        full_history=full_history,
        short_history=short_history,
        topic=topic,
        subject=subject,
        last_class=last_class,
        last_subject=last_subject,
    )


def _display_name_from_timestamp(ts: float) -> str:
    from datetime import datetime

    dt = datetime.fromtimestamp(ts)
    return dt.strftime("Chat – %b %d, %I:%M %p")


def summarize_sessions() -> List[Dict[str, Any]]:
    if not HISTORY_FILE.exists():
        return []
    sessions: Dict[str, Dict[str, Any]] = {}
    with HISTORY_FILE.open(encoding="utf-8") as handle:
        for line in handle:
            try:
                record = json.loads(line)
            except json.JSONDecodeError:
                continue
            session_id = record.get("session_id")
            ts = record.get("timestamp")
            if not session_id or not ts:
                continue
            session = sessions.setdefault(
                session_id,
                {"session_id": session_id, "created_at": ts, "updated_at": ts},
            )
            session["created_at"] = min(session["created_at"], ts)
            session["updated_at"] = max(session["updated_at"], ts)
    for session in sessions.values():
        session["title"] = _display_name_from_timestamp(session["created_at"])
    ordered = sorted(sessions.values(), key=lambda item: item["updated_at"], reverse=True)
    return ordered


def get_session_summary(session_id: str) -> Optional[Dict[str, Any]]:
    for session in summarize_sessions():
        if session["session_id"] == session_id:
            return session
    return None


def describe_providers() -> Dict[str, Dict[str, Any]]:
    """Single local provider for vanilla kiosk."""
    from .msai_bridge import DEFAULT_CHECKPOINT

    configured = DEFAULT_CHECKPOINT.exists()
    note = "" if configured else f"Place best.pt at {DEFAULT_CHECKPOINT}"
    return {
        "vanilla": {
            "id": "vanilla",
            "name": "Vanilla tool-calling LM (local)",
            "configured": configured,
            "default_model": "decoder-only",
            "note": note,
        }
    }


@lru_cache(maxsize=1)
def get_orchestrator() -> ConversationOrchestrator:
    warm_load()
    catalog = load_default_catalog(ARCHIVE_DIR)
    engine = AnalysisEngine(
        catalog,
        [
            FacultyByTopicBlueprint(),
            LocationBlueprint(),
            CenterBlueprint(),
            AdvisorshipBlueprint(),
            StaffSupportBlueprint(),
            UpcomingEventsBlueprint(),
            OfficeHoursBlueprint(),
            PersonLookupBlueprint(),
        ],
    )
    try:
        engine.refresh_events()
    except Exception:
        pass
    return ConversationOrchestrator(engine)


@app.get("/api/providers")
def providers_endpoint() -> Dict[str, Any]:
    inventory = describe_providers()
    return {"providers": inventory, "default_provider": "vanilla"}


@app.post("/api/query")
def query(payload: QueryPayload) -> Dict[str, Any]:
    question = (payload.question or "").strip()
    if not question:
        raise HTTPException(status_code=400, detail="Question is required.")

    session_id = (payload.session_id or DEFAULT_SESSION).strip() or DEFAULT_SESSION
    requested_provider = (payload.provider or "").strip().lower() or None
    canonical_provider = normalize_provider_name(requested_provider) if requested_provider else None

    if canonical_provider and canonical_provider != "vanilla":
        raise HTTPException(
            status_code=400,
            detail="This kiosk only supports the local vanilla LM. Omit provider or use provider=vanilla.",
        )
    inventory = describe_providers()
    vanilla_meta = inventory.get("vanilla", {})
    if not vanilla_meta.get("configured", False):
        note = vanilla_meta.get("note") or "Vanilla checkpoint not found under kiosk_vanilla/models/."
        raise HTTPException(status_code=503, detail=note)

    history = load_history(session_id)
    planner_context = build_planner_context_from_history(history)
    planner_context = strip_context_on_topic_switch(question, planner_context)
    resolved = resolve_context(question, planner_context)

    orchestrator = get_orchestrator()
    with _orchestrator_lock:
        orchestrator.ensure_responder(canonical_provider)
        answer, result, action = orchestrator.answer(
            resolved.question,
            context=planner_context,
            resolved_input=resolved,
        )
        metadata = (
            orchestrator.responder.get_metadata()
            if hasattr(orchestrator, "responder") and hasattr(orchestrator.responder, "get_metadata")
            else {}
        )
        metadata.setdefault("planner_action", action.to_dict())

    facts_payload = [fact.__dict__ for fact in result.facts]
    record_history(
        session_id=session_id,
        question=question,
        answer=answer,
        blueprint=result.name,
        metadata=metadata,
        facts=facts_payload,
        notes=result.notes,
        action=action.to_dict(),
    )
    summary = get_session_summary(session_id) or {
        "session_id": session_id,
        "title": _display_name_from_timestamp(time.time()),
    }

    return {
        "session_id": session_id,
        "session_title": summary.get("title"),
        "question": question,
        "answer": answer,
        "blueprint": result.name,
        "facts": facts_payload,
        "notes": result.notes,
        "usage": metadata,
        "action": action.to_dict(),
    }


@app.get("/api/history")
def history(session_id: str = Query(DEFAULT_SESSION)) -> Dict[str, Any]:
    entries = load_history(session_id)
    summary = get_session_summary(session_id)
    title = summary.get("title") if summary else _display_name_from_timestamp(time.time())
    return {"session_id": session_id, "title": title, "history": entries}


@app.get("/api/sessions")
def sessions() -> Dict[str, Any]:
    return {"sessions": summarize_sessions()}


def _register_frontend_routes() -> None:
    # If there is no built frontend, fall back to a dev-server redirect so
    # developers running `npm run dev` can visit the backend host and be
    # forwarded to Vite (or another dev server). The dev URL may be set via
    # `KIOSK_FRONTEND_DEV_URL` (defaults to http://127.0.0.1:5173).
    if not FRONTEND_DIST.exists():
        dev_url = os.getenv("KIOSK_FRONTEND_DEV_URL", "http://127.0.0.1:5173").rstrip("/")

        @app.get("/", include_in_schema=False)
        async def dev_redirect_root() -> RedirectResponse:  # pragma: no cover - dev helper
            return RedirectResponse(dev_url)

        @app.get("/{full_path:path}", include_in_schema=False)
        async def dev_redirect(full_path: str) -> RedirectResponse:  # pragma: no cover - dev helper
            if full_path.startswith("api/"):
                raise HTTPException(status_code=404, detail="Unknown API route.")
            if not full_path:
                return RedirectResponse(dev_url)
            target = f"{dev_url}/{full_path.lstrip('/')}"
            return RedirectResponse(target)

        return

    assets_dir = FRONTEND_DIST / "assets"
    if assets_dir.exists():
        app.mount("/assets", StaticFiles(directory=assets_dir), name="frontend-assets")

    def _serve_frontend() -> FileResponse:
        if FRONTEND_INDEX.exists():
            return FileResponse(FRONTEND_INDEX)
        raise HTTPException(status_code=404, detail="Frontend build is missing.")

    @app.get("/", include_in_schema=False)
    async def serve_frontend_root() -> FileResponse:  # pragma: no cover - deployment wiring
        return _serve_frontend()

    @app.get("/{full_path:path}", include_in_schema=False)
    async def serve_frontend_routes(full_path: str) -> FileResponse:  # pragma: no cover - deployment wiring
        if full_path.startswith("api/"):
            raise HTTPException(status_code=404, detail="Unknown API route.")
        return _serve_frontend()


_register_frontend_routes()


def main() -> None:
    import uvicorn

    uvicorn.run(
        "backend.main:app",
        host="127.0.0.1",
        port=5050,
        reload=False,
    )


if __name__ == "__main__":
    main()