Kavach Security Architecture Overview
Kavach defends modern software organizations by combining deterministic policy guardrails, isolated scanner workers, and runtime eBPF telemetry. Here is how our architecture protects customer assets and data.
In our model, artificial intelligence cannot make unilateral security decisions. Critical merge gates, cloud configuration changes, and secret revocations are enforced through Open Policy Agent (OPA) Rego rules, scanner outputs, and human verification.
Cryptographic Data Isolation
All network data in transit is encrypted using modern TLS 1.3 with forward secrecy. PostgreSQL customer telemetry and policy versions are encrypted at rest using AES-256 with KMS keys.
Worker Micro-Sandboxes
Repositories cloned for static analysis (Semgrep, Gitleaks, Trivy) run inside memory-isolated worker sandboxes with restricted network egress. All source files are wiped immediately upon scan termination.
Secret Redaction in Audit
Kavach's MCPAuditor intercepts every tool input and output payload, automatically redacting API keys, bearer tokens, and private credentials to **** before persisting audit logs.
Prohibited Dangerous Permissions
Autonomous AI agents connected via MCP are strictly prohibited from receiving destructive system rights like EXECUTE_SHELL, SECRET_READ, or DATABASE_DELETE.
Vulnerability Reporting & Security Contact
To report suspected security incidents or vulnerabilities in our infrastructure, please contact our Security Engineering Response Team:
/.well-known/security.txt.