Kern

•
13
•

Verifies AI-generated code and blocks unsafe pull requests using formal SMT proofs, four-valued Belnap logic, and automated anti-cheat detection.

Kern operates as an L0–L5 zero-trust verification gateway positioned between AI coding agents and production repositories, preventing flawed or deceptive code from slipping through continuous integration. While standard CI pipelines only measure exit codes and often pass vacuous tests or hallucinated fixes, Kern leverages Z3 SMT solving, Lean 4 theorems, and a four-valued Belnap B₄ logic lattice to mathematically prove contracts across the entire domain or produce concrete counterexamples. Written in high-performance Zig, it detects common agent test-cheating techniques—such as swallowed assertions and bypassed tests—and exposes its verification capabilities via native C-ABIs, HTTP guardrails, and an MCP server interface.

Key Features

01Can-fail harness discipline that ensures checks cannot pass vacuously via fault injection
02Four-valued Belnap B4 logic lattice that isolates conflicts and maps directly to RFC 9110 HTTP statuses
03Native Z3 proof-DAG verification and formal verification backed by Lean 4 theorems
04Automated detection of AI agent test cheating, including swallowed exceptions and mocked assertions
05High-performance Zig implementation featuring SIMD-accelerated BLAKE3 hashing and an MCP server interface
0613 GitHub stars

Use Cases

01Gating AI coding agent output to ensure generated code and contracts mathematically hold across edge cases
02Preventing test circumvention and prompt injection tricks within automated agentic workflows and RLVR pipelines
03Resolving cloud IAM and financial logic conflicts by detecting contradictory policies rather than silently masking errors