Your codebase stays on your machine. Kernax combines local AST parsing with stateless, zero-training API routing so you never trade intellectual property for autonomous speed.
Kernax operates as a pass-through orchestration layer, not a code warehouse. We do not clone your repositories to remote servers or store your source files in persistent databases.
Codebase mapping happens locally on your device via Tree-sitter AST graphs. Only the exact AST nodes and file slices required to solve your active task traverse our encrypted API gateway—keeping raw API credentials off developer machines.
Kernax routes inference workloads through vetted commercial API gateways—including Fireworks AI and OpenRouter—strictly configured to deny prompt training. Neither Kernax nor our inference providers train foundation models on your code.
Autonomous agents shouldn't mean unchecked shell access. Kernax enforces kernel-level command interception and granular tool permissions directly inside your local environment before a single line executes.
Engineered around strict data minimization, transport encryption, and transparent legal accountability.
All traffic between the Kernax client, our gateway, and inference APIs is encrypted in transit via Cloudflare TLS 1.3.
Commercial API payloads are contractually and programmatically excluded from model weights and dataset training.
Engineered under strict data minimization principles—collecting only billing and account metadata with zero telemetry on source code.