# CVE Fact Checker Flask API

Production-oriented API layer exposing fact checking, Firebase ingestion, and semantic search.

## Features
- Firebase article ingestion -> vector store (Chroma)
- Semantic similarity search
- Claim fact-check endpoint (LLM backed if OPENROUTER_API_KEY present; graceful fallback otherwise)
- Schema inspection & collection listing
- Lightweight dummy embeddings fallback (no GPU / HF model required)

## Environment Variables
```
OPENROUTER_API_KEY=your_key           # Optional (improves analysis)
APP_DOMAIN=https://yourdomain.tld     # For OpenRouter headers (optional)
APP_TITLE="Your App Title"            # For OpenRouter headers (optional)
```

## Install
```
python -m venv .venv
.venv\Scripts\activate  # Windows
pip install -r requirements.txt
pip install flask
```

## Run
```
python -m cve_factchecker.app
```
Server listens on `http://0.0.0.0:8000`.

## Endpoints
| Method | Path | Description |
| ------ | ---- | ----------- |
| GET | / | Index + endpoint list |
| GET | /health | Health & uptime |
| POST | /ingest/firebase | Ingest from Firebase (paged) `{collection?, limit?}` |
| POST | /ingest/firebase/full | Ingest ALL documents (pagination) `{collection?}` |
| GET | /schema | Inspect collection schema `?collection=Articles` |
| GET | /collections | List probable collections |
| GET | /search | Semantic search `?q=term&k=5` |
| POST | /fact-check | Fact check `{claim, auto_ingest?, ingest_limit?}` |

## Example Requests (PowerShell)
```
# Health
curl http://localhost:8000/health

# Ingest sample (limited)
curl -Method POST -Uri http://localhost:8000/ingest/firebase -Headers @{"Content-Type"="application/json"} -Body '{"limit":50}'

# Full ingest (all)
curl -Method POST -Uri http://localhost:8000/ingest/firebase/full -Headers @{"Content-Type"="application/json"} -Body '{}'

# Search
curl "http://localhost:8000/search?q=Pakistan"

# Fact Check (with auto ingest)
curl -Method POST -Uri http://localhost:8000/fact-check -Headers @{"Content-Type"="application/json"} -Body '{"claim":"Pakistan signed a new mineral deal","auto_ingest":true,"ingest_limit":80}'
```

## Notes
- If you have no API key configured, verdicts will be `UNVERIFIED`; retrieval still works.
- Adjust vector DB path by editing `FactCheckSystem` init in `orchestrator.py`.
- Set `AUTO_INGEST_FIREBASE=true` (env var) to automatically ingest all articles on startup.