config.py

"""
Central config for the JanMat RAG chatbot backend.

Everything secret comes from the environment (HF Space repository secrets in
prod, .env locally — see .env.example). Nothing here should hold a real key.
"""
import os
from dotenv import load_dotenv

load_dotenv()  # harmless no-op if you never create a .env file

import os
from dotenv import load_dotenv

load_dotenv()  # harmless no-op if you never create a .env file — see note below

# ==============================================================================
# All secrets are read from environment variables, NEVER hardcoded here.
# This file is safe to commit to GitHub (public or private) as-is.
#
# Where the environment variables actually come from depends on where you run:
#   - Locally: `export SUPABASE_URL="..."` etc. in your terminal before running
#     `python main.py` (see MIGRATION_GUIDE.md for the exact commands).
#   - Hugging Face Space: Settings -> Variables and secrets -> add each one
#     under "Secrets". HF injects them as environment variables automatically
#     at runtime — nothing in this file needs to change.
# ==============================================================================

# --- Supabase (knowledge base + chat history) --------------------------------
SUPABASE_URL = os.getenv("SUPABASE_URL", "")
SUPABASE_SERVICE_KEY = os.getenv("SUPABASE_SERVICE_KEY", "")  # service_role key, backend-only

# --- Groq (answer generation) -------------------------------------------------
GROQ_API_KEY = os.getenv("GROQ_API_KEY", "")
GROQ_CHAT_MODEL = os.getenv("GROQ_CHAT_MODEL", "openai/gpt-oss-20b")

# --- Sarvam AI (voice: speech-to-text + text-to-speech) -----------------------
SARVAM_API_KEY = os.getenv("SARVAM_API_KEY", "")
SARVAM_STT_MODEL = os.getenv("SARVAM_STT_MODEL", "saaras:v3")
SARVAM_TTS_MODEL = os.getenv("SARVAM_TTS_MODEL", "bulbul:v3")
SARVAM_DEFAULT_SPEAKER = os.getenv("SARVAM_DEFAULT_SPEAKER", "anushka")
SARVAM_DEFAULT_LANGUAGE = os.getenv("SARVAM_DEFAULT_LANGUAGE", "en-IN")

# --- TinyFish (live web: Search, Fetch, Agent) -------------------------------
# Optional. Leave unset and JanMat behaves exactly as before.
TINYFISH_API_KEY = os.getenv("TINYFISH_API_KEY", "")

# --- Cron / admin protection ---------------------------------------------------
CRON_SECRET = os.getenv("CRON_SECRET", "")

# --- Embedding model (local, free, no API key needed) -------------------------
EMBEDDING_MODEL_NAME = os.getenv("EMBEDDING_MODEL_NAME", "sentence-transformers/all-MiniLM-L6-v2")
EMBEDDING_DIM = 384

# --- Chunking -------------------------------------------------------------------
CHUNK_WORDS = int(os.getenv("CHUNK_WORDS", "220"))
CHUNK_OVERLAP_WORDS = int(os.getenv("CHUNK_OVERLAP_WORDS", "40"))

# --- Retrieval / generation -------------------------------------------------------
# Broader than a typical minimal RAG setup: more candidate chunks per query,
# and a larger context budget, so answers draw on more of the ingested
# material per question rather than a bare top-5. Still cheap — this is a
# few dozen KB of text per Groq call, not the whole knowledge base.
TOP_K_CHUNKS = int(os.getenv("TOP_K_CHUNKS", "20"))
MAX_CONTEXT_CHARS = int(os.getenv("MAX_CONTEXT_CHARS", "20000"))

# --- Sources to ingest ------------------------------------------------------------
CONFIG = {
    "prs": {
        "name": "PRS Legislative Research",
        "base_url": "https://prsindia.org",
        "listing_url": "https://prsindia.org/billtrack",
        "bill_link_pattern": r"^/billtrack/[^/]+$",
        "bill_link_exclude_prefixes": ("/billtrack/category/", "/billtrack/field_bill_category/"),
        "rate_limit_seconds": 2,
        # Rolling window: JanMat keeps the newest N bills (PRS lists newest first). Each ingestion
        # adds any new bills and then deletes the oldest ones beyond N, with their PDFs and text.
        "max_bills_kept": 25,
        "listing_pages": 6,                  # how many ?page=N listing pages to read to find N bills
        "delete_comments_with_bill": False,  # True = also delete user comments on bills that rotate out
        "license_note": (
            "PRS's disclaimer permits reproduction for non-commercial purposes "
            "with due acknowledgement of PRS Legislative Research."
        ),
    },
    # The rest of the PRS website (everything beyond the first page of bills).
    # The crawler visits each section's index pages (following ?page=N pagination),
    # collects links whose path matches `item_pattern`, and ingests each page plus
    # any PDFs it links to. Add or remove sections freely.
    "prs_library": {
        "enabled": True,
        # Each run ingests at most this many NEW documents, then stops. Documents
        # already stored are skipped without being fetched, so just click
        # "Run ingestion now" repeatedly and the library grows until it is complete.
        # (Free Supabase = 500 MB; roughly 4 KB per chunk, ~100 chunks per document.)
        "max_new_docs_per_run": 60,
        "max_index_pages": 5,        # pagination depth per index URL
        "sections": [
            {"key": "policy_review", "label": "Policy Reviews",
             "index_urls": ["https://prsindia.org/policy/monthly-policy-review",
                            "https://prsindia.org/policy/annual-policy-review"],
             "item_pattern": r"^/policy/(monthly|annual)-policy-review/[^/]+$"},
            {"key": "committee_report", "label": "Committee Reports",
             "index_urls": ["https://prsindia.org/policy/report-summaries",
                            "https://prsindia.org/parliamenttrack/report-summaries"],
             "item_pattern": r"^/(policy|parliamenttrack)/report-summaries/[^/]+$"},
            {"key": "session", "label": "Session Wraps",
             "index_urls": ["https://prsindia.org/sessiontrack"],
             "item_pattern": r"^/sessiontrack/[^/]+(/[^/]+)?$"},
            {"key": "discussion_paper", "label": "Discussion Papers",
             "index_urls": ["https://prsindia.org/policy/discussion-papers",
                            "https://prsindia.org/legislatures/discussion-papers",
                            "https://prsindia.org/budgets/discussionpapers"],
             "item_pattern": r"^/(policy|legislatures|budgets)/(discussion-papers|discussionpapers)/[^/]+$"},
            {"key": "blog", "label": "Blog",
             "index_urls": ["https://prsindia.org/theprsblog"],
             "item_pattern": r"^/theprsblog/(?!blog-author)[^/]+$"},
            {"key": "article", "label": "Articles",
             "index_urls": ["https://prsindia.org/articles-by-prs-team/2026",
                            "https://prsindia.org/articles-by-prs-team/2025",
                            "https://prsindia.org/articles-by-prs-team/2024"],
             "item_pattern": r"^/articles-by-prs-team/(?!\d{4}$)[^/]+$"},
            {"key": "act", "label": "Acts",
             "index_urls": ["https://prsindia.org/acts/parliament", "https://prsindia.org/recent-acts"],
             "item_pattern": r"^/(acts/[^/]+|recent-acts)/[^/]+$"},
            {"key": "state_bill", "label": "State Bills",
             "index_urls": ["https://prsindia.org/bills/states",
                            "https://prsindia.org/bills/state-legislative-briefs"],
             "item_pattern": r"^/bills/(states|state-legislative-briefs)/[^/]+$"},
            {"key": "budget", "label": "Budgets",
             "index_urls": ["https://prsindia.org/budgets/parliament", "https://prsindia.org/budgets/states"],
             "item_pattern": r"^/budgets/(parliament|states)/[^/]+$"},
        ],
        # "Comments invited on ..." notices (bill name, deadline, link to submit).
        "announcement_urls": ["https://prsindia.org/", "https://prsindia.org/announcements"],
    },
    # Add more sources here as plain "seed URL lists" — the ingestion pipeline
    # treats any HTTP(S) URL the same way: fetch, detect PDF vs HTML, extract
    # full text, chunk, embed, store. See scraper.py:ArticleSource.
    # Used only when TINYFISH_API_KEY is set.
    "tinyfish": {
        # Search API: finds drafts / news / stakeholder reactions beyond PRS.
        "search_queries": [
            "draft bill public consultation stakeholder comments India",
            "Ministry of Corporate Affairs draft amendment comments invited",
        ],
        "results_per_query": 5,
        # Agent API: pages whose public comments should be extracted. Add the
        # consultation pages you are allowed to read (e.g. MCA draft pages).
        "consultation_urls": [],
        # Also run the agent on this many PRS bill pages per run (Agent runs are billed).
        "agent_max_bill_pages": 3,
    },
    "article_seed_urls": [
        # "https://example-news-site.com/article-about-a-bill",
    ],
}