Home
Stop documentation drift before it reaches production.
AST-aware drift detection, AI severity triage, and agent-driven remediation for enterprise codebases.
Detection
Tree-sitter AST fingerprinting compares code structure, not text. Ignores whitespace and comment changes. Tracks specific methods and classes. Survives rebase and squash.
Triage
A lightweight LLM call classifies each drift as no_update, minor, or major. No AI cost for clean docs. Triage runs only on drifted anchors.
Remediation
Invokes your agent (Kiro, Claude Code, or custom) to update docs, stamp new provenance, and open merge requests. kedge's pipeline ends when the agent returns — MR review and merging follow your existing workflows.
Enterprise CI native
Runs as an MR gate (kedge check, exit 1 on drift) or a scheduled pipeline (kedge update for full detect-triage-remediate). The repo includes GitLab CI and GitHub Actions examples.
Multi-language, multi-team
AST fingerprinting for Java, Go, TypeScript, Python, Rust, and XML. Content-hash fallback for everything else. Group-scoped steering files let each team own their docs.
Zero runtime dependencies
Single static binary. No JVM, no Node, no Python. Uses rustls (no OpenSSL) and git CLI (no libgit2 at runtime). Copy it to your CI runner and go.