How chirag0071/saffronai---agentic-honey-pot-for-scam-detection works
SAFFRON AI is a FastAPI scam-honeypot backend: GUVI or another caller submits one message per session, and the API retrieves conversation state, generates a human-like reply, extracts scam intelligence, and sends a final GUVI report when the session meets its completion threshold. The runtime includes engagement, forensic extraction, attack-flow and risk analysis, and callback reporting. The session store is in-memory. The README documents these product capabilities; the graph’s implementation-level links are limited to the provided excerpts and documented flow.
Main entry points: intelligence.py Intelligence extractor · main.py Message API · models.py Session state · session_manager.py Session manager
Components 15
API and sessions
- Message API
backend/app/api/main.py - Session manager
backend/app/core/session_manager.py - Session state
backend/app/models/models.py
Scammer engagement
- Honeypot agent
backend/app/core/agent.py - LLM client
backend/app/services/llm_client.py
Intelligence analysis
- Intelligence extractor
backend/app/core/intelligence.py - Extracted indicators
backend/app/core/intelligence.py - Scam classification
backend/app/core/intelligence.py - Confidence scoring
backend/app/core/intelligence.py - Attack reconstruction
backend/app/core/intelligence.py - Risk estimation
backend/app/core/intelligence.py - Completion threshold
backend/app/core/intelligence.py
Result reporting
- GUVI callback client
backend/app/services/guvi_client.py
Other components
- Message caller
- GUVI evaluator
Connections 17
- Message caller → Message API: submits message
README.md - Message API → Session manager: gets session
backend/app/api/main.py - Session manager → Session state: creates state
backend/app/core/session_manager.py - Message API → Session state: appends messages
backend/app/api/main.py - Message API → Honeypot agent: requests reply
backend/app/api/main.py - Honeypot agent → LLM client: generates reply
backend/app/core/agent.py - Message API → Intelligence extractor: extracts intelligence
backend/app/api/main.py - Intelligence extractor → Session state: reads and updates
backend/app/core/intelligence.py - Intelligence extractor → LLM client: requests extraction
backend/app/core/intelligence.py - Intelligence extractor → Extracted indicators: extracts indicators
backend/app/core/intelligence.py - Intelligence extractor → Scam classification: classifies scam
backend/app/core/intelligence.py - Intelligence extractor → Confidence scoring: scores confidence
backend/app/core/intelligence.py - Intelligence extractor → Attack reconstruction: infers attack flow
backend/app/core/intelligence.py - Intelligence extractor → Risk estimation: calculates risk
backend/app/core/intelligence.py - Intelligence extractor → Completion threshold: checks threshold
backend/app/core/intelligence.py - Message API → GUVI callback client: sends final result
backend/app/api/main.py - GUVI callback client → GUVI evaluator: posts report
backend/app/services/guvi_client.py