How harshsatbhai/guardrailx works
This is a prompt-injection and jailbreak guardrail with a React chat/dashboard, Express orchestration and security controls, a Python ML service, and persistent logging. The main workflow classifies text or extracted media, blocks or warns by risk, sends allowed prompts to a selectable LLM, filters its response, and logs the outcome. Authenticated users can also manage chat sessions and view security analytics. Relationships are limited to those shown in excerpts, the source index, or README; several UI and route details are documented but not sampled.
Main entry points: server.js Express server · App.jsx App routing · pipelineShared.js Post-classification · api.js Backend API client
Components 30
Web client
- App routing
frontend/src/App.jsx - Auth UI
frontend/src/pages/AuthPage.jsx - Auth state
frontend/src/context/AuthContext.jsx - Chat page
frontend/src/pages/ChatPage.jsx - Chat rendering
frontend/src/components/ChatMessage.jsx - Voice input
frontend/src/hooks/useVoiceInput.js - Chat sessions
frontend/src/hooks/useChatSessions.js - Dashboard page
frontend/src/pages/DashboardPage.jsx - Backend API client
frontend/src/api.js
Backend and security
- Express server
backend/src/server.js - Auth routes
backend/src/routes/auth.js - Chat routes
backend/src/routes/chats.js - Prompt routes
backend/src/routes/checkPrompt.js - Media routes
backend/src/routes/checkMultimodal.js - Post-classification
backend/src/routes/pipelineShared.js - Rate and ban controls
backend/src/middleware/rateLimiter.js - ML service client
backend/src/services/mlService.js - Dashboard analytics
backend/src/routes/dashboard.js
ML guardrails
- ML API
ml-service/app/main.py - Media extraction
ml-service/app/multimodal.py - Risk classifier
ml-service/app/classify.py - NLP and features
ml-service/app/features.py - Response redaction
ml-service/app/response_filter.py
State and providers
- Prompt logs
backend/src/models/PromptLog.js - Chat sessions store
backend/src/models/ChatSession.js - User store
backend/src/models/User.js - IP blacklist
backend/src/models/IpBlacklist.js - MongoDB
- LLM provider
Other components
- User
Connections 41
- User → App routing: uses
README.md - App routing → Chat page: routes to
frontend/src/App.jsx - App routing → Dashboard page: routes to
frontend/src/App.jsx - App routing → Auth UI: routes to
frontend/src/App.jsx - App routing → Auth state: provides auth
frontend/src/App.jsx - App routing → Chat sessions: loads sessions
frontend/src/App.jsx - Chat page → Backend API client: requests
frontend/src/pages/ChatPage.jsx - Chat page → Chat rendering: renders
frontend/src/pages/ChatPage.jsx - Chat page → Voice input: uses
frontend/src/pages/ChatPage.jsx - Chat sessions → Backend API client: loads and saves
frontend/src/hooks/useChatSessions.js - Dashboard page → Backend API client: requests
frontend/src/pages/DashboardPage.jsx - Auth state → Backend API client: authenticates
frontend/src/context/AuthContext.jsx - Backend API client → Express server: calls API
README.md - Express server → Auth routes: mounts
backend/src/server.js - Express server → Chat routes: mounts
backend/src/server.js - Express server → Prompt routes: mounts
backend/src/server.js - Express server → Media routes: mounts
backend/src/server.js - Express server → Dashboard analytics: mounts
backend/src/server.js - Express server → Rate and ban controls: guards prompt routes
backend/src/server.js - Prompt routes → ML service client: classifies prompt
README.md - Media routes → ML service client: classifies media
backend/src/routes/checkMultimodal.js - Prompt routes → Post-classification: continues pipeline
backend/src/routes/pipelineShared.js - Media routes → Post-classification: continues pipeline
backend/src/routes/checkMultimodal.js - Post-classification → ML service client: filters response
backend/src/routes/pipelineShared.js - Post-classification → LLM provider: generates allowed response
backend/src/routes/pipelineShared.js - Post-classification → Rate and ban controls: records violation
backend/src/routes/pipelineShared.js - Post-classification → Prompt logs: writes outcome
backend/src/routes/pipelineShared.js - ML service client → ML API: HTTP requests
backend/src/services/mlService.js - ML API → Risk classifier: classifies text
ml-service/app/main.py - ML API → Response redaction: filters output
ml-service/app/main.py - ML API → Media extraction: extracts media text
ml-service/app/main.py - Media extraction → Risk classifier: sends extracted text
ml-service/app/main.py - Risk classifier → NLP and features: uses features
ml-service/app/classify.py - Auth routes → User store: uses accounts
backend/src/routes/auth.js - Chat routes → Chat sessions store: uses sessions
backend/src/routes/chats.js - Dashboard analytics → Prompt logs: reads analytics
backend/src/routes/dashboard.js - Dashboard analytics → IP blacklist: manages bans
backend/src/routes/dashboard.js - Prompt logs → MongoDB: persists in
- Chat sessions store → MongoDB: persists in
- User store → MongoDB: persists in
- IP blacklist → MongoDB: persists in