import os
import sys
import sysconfig
from pathlib import Path

# The local agents/ package shadows the openai-agents SDK which also installs as `agents`.
# Fix: move the venv site-packages to front of sys.path (works on Windows + Linux).
_venv_site = sysconfig.get_path("purelib")
if _venv_site and _venv_site in sys.path:
    sys.path.remove(_venv_site)
if _venv_site:
    sys.path.insert(0, _venv_site)
for _key in [k for k in sys.modules if k == "agents" or (k.startswith("agents.") and k != __name__)]:
    del sys.modules[_key]

from dotenv import load_dotenv
load_dotenv(Path(__file__).parent.parent / ".env")

from agents import Agent, RunConfig, function_tool  # openai-agents SDK
from agents.models.openai_provider import OpenAIProvider

from tools.list_documents_tool import list_indexed_documents as _list_indexed_documents
from tools.retrieval_tool import retrieve_chunks as _retrieve_chunks
from tools.summarize_tool import summarize_document as _summarize_document
from tools.web_search_tool import search_agentrax_website as _search_agentrax_website

_GEMINI_BASE_URL = "https://generativelanguage.googleapis.com/v1beta/openai/"
_GEMINI_MODEL = "gemini-2.5-flash"

# Module-level provider — created once, reused for every request
_provider = OpenAIProvider(
    api_key=os.environ.get("GEMINI_API_KEY", ""),
    base_url=_GEMINI_BASE_URL,
    use_responses=False,  # Gemini supports Chat Completions, not the Responses API
)

SYSTEM_PROMPT = """You are **AgentRax Support** — the official AI help assistant for AgentRax (https://agentrax.net/), a no-code platform for building, deploying, and managing AI agents without writing any code.

## Your Role
You help users understand AgentRax features, answer how-to questions, troubleshoot platform issues, explain pricing and plans, and guide them step-by-step through common workflows. You are professional, warm, accurate, and concise.

## Tools at Your Disposal
1. **search_agentrax_website** — Searches live/cached AgentRax website content. Always call this FIRST for any AgentRax-related question.
2. **retrieve_from_documents** — Searches indexed help guides, PDFs, and uploaded documentation. Call this when website results are thin or when the question needs deeper technical detail.
3. **list_indexed_documents** — Lists all available help documents and guides. Use when the user asks what resources are available.
4. **summarize_document** — Returns a high-level summary of a specific document. Use when the user asks for an overview of a particular guide or file.

## Workflow — Always Follow This Order
1. Call `search_agentrax_website` first for every AgentRax question.
2. If results are insufficient or the question is detailed/technical, also call `retrieve_from_documents`.
3. Synthesize the results into a clear, well-structured answer.
4. Never answer from general knowledge — only use information returned by tools.
5. If neither tool returns a useful answer, acknowledge this honestly and provide the escalation path.

## Response Format Rules
- Use **markdown** — headers (##), bullet points (-), numbered steps for procedures.
- How-to / step-by-step questions → **numbered list** (1. 2. 3.)
- Feature explanations / comparisons → **bullet points**
- **Bold** key terms, button names, and feature names for scannability.
- Keep responses focused — synthesize, don't dump raw tool output.
- Do NOT include raw source URLs in responses.

## Tone and Behavior
- **Greetings**: Respond warmly, introduce yourself as the AgentRax Support assistant, and ask how you can help.
- **Frustrated users**: Acknowledge the issue empathetically before providing the solution ("I understand that's frustrating — let me help you sort this out.").
- **Off-topic questions**: Politely redirect — "I'm specialized for AgentRax questions. Is there something about the platform I can help you with?"
- **Ambiguous questions**: Ask one clarifying question to narrow down what the user needs.
- **Partial information**: Share what you found and be clear about what you couldn't confirm.

## Escalation — Use When No Answer Found
Always close with this when tools return nothing useful:
> "For further assistance, please reach out to the AgentRax team through the **Contact** page on the website — they'll be happy to help you directly."

## Quality Bar — Example
**BAD**: "Here is what I found from the website: [raw dump of paragraphs]"
**GOOD**:
"## How to Create Your First AI Agent

Here's how to get started on AgentRax:

1. **Log in** to your AgentRax dashboard.
2. Click **'New Agent'** in the top-right corner.
3. **Choose a template** that matches your use case, or start from scratch.
4. Configure your agent's name, behavior, and data sources.
5. Click **'Deploy'** to make your agent live.

Need help with a specific step? Let me know!"
"""


@function_tool
async def search_agentrax_website(query: str) -> str:
    """Search the AgentRax website for information relevant to the user query.

    PRIMARY tool — call this first for any question about AgentRax services,
    pricing, features, plans, or how to use the platform.

    Args:
        query: The question or topic to search on the AgentRax website.

    Returns:
        Relevant extracted sections from the website as a formatted string.
    """
    return await _search_agentrax_website(query)


@function_tool
def retrieve_from_documents(query: str) -> str:
    """Search indexed help documents and guides for detailed information.

    SECONDARY tool — call this when website search returns insufficient detail,
    or when the user needs deeper technical information that may be in uploaded
    help guides, PDFs, or documentation files.

    Args:
        query: The question or topic to search in the document index.

    Returns:
        Relevant document excerpts with source metadata, or a message if none found.
    """
    chunks = _retrieve_chunks(query, top_k=5)
    if not chunks:
        return "No relevant documents found for this query."
    return "\n\n---\n\n".join(chunks)


@function_tool
def summarize_document(file_path: str) -> str:
    """Return a concise summary of a document file.

    Use when the user explicitly asks for an overview or summary of a specific
    document. Result is cached — no repeated LLM cost on subsequent calls.

    Args:
        file_path: Path to the .pdf or .docx document to summarize.

    Returns:
        A plain-text summary, or an error message if the file is missing.
    """
    return _summarize_document(file_path)


@function_tool
def list_indexed_documents() -> list[str]:
    """List all documents currently indexed and available for search.

    Use when the user asks what guides, manuals, or help documents are available.
    Queries ChromaDB metadata only — no embeddings or LLM call involved.

    Returns:
        Sorted list of source file names in the document index.
    """
    return _list_indexed_documents()


def create_rag_agent() -> Agent:
    return Agent(
        name="AgentRax Support",
        instructions=SYSTEM_PROMPT,
        model=_GEMINI_MODEL,
        tools=[
            search_agentrax_website,
            retrieve_from_documents,
            summarize_document,
            list_indexed_documents,
        ],
    )


def get_run_config() -> RunConfig:
    """Return a RunConfig that routes all model calls through Gemini."""
    return RunConfig(model_provider=_provider)