Welcome to the LLMTrace documentation. LLMTrace is a transparent proxy that captures, analyses, and secures LLM interactions in real-time.
- Installation -- build from source, Docker, or Kubernetes
- Quick Start -- get running in under 5 minutes
- Configuration -- config file reference
- OpenAI SDK Integration -- Python and Node.js SDK setup
- LangChain Integration -- LangChain/LangGraph integration
- curl / HTTP Integration -- raw HTTP and scripting examples
- Custom Security Policies -- define and deploy custom detection rules
- Pre-Request Enforcement -- block, flag, or log requests before forwarding
- Auth & Multi-Tenancy -- API keys, RBAC roles, tenant isolation
- Dashboard -- built-in Next.js dashboard usage
- Monitoring & Observability -- Prometheus metrics, health checks, alerting
- Troubleshooting -- common issues and diagnostic steps
- Integration Tests -- testing your LLMTrace setup
- ML Model Reference -- DeBERTa, InjecGuard, PIGuard, jailbreak detector
- Ensemble Detection -- how multi-detector voting works
- Threshold Tuning -- operating points, per-category thresholds, over-defence
- Benchmark Methodology -- test corpus, results, reproduction steps
- OWASP LLM Top 10 -- coverage mapping
- Kubernetes -- Helm charts, manifests, HA setup
- Secrets Management -- Vault, KMS, environment variables
- System Architecture -- component overview and data flow
- Transparent Proxy -- how the proxy intercepts traffic
- REST API -- endpoint reference
Swagger UI: -- available at /swagger-ui when the proxy is running
The examples/ directory contains ready-to-use configuration files and integration scripts:
config.example.yaml-- minimal proxy configconfig.production.yaml-- production-grade configpython/-- Python SDK examplesnode/-- Node.js examplescurl/-- shell scripts for testing