Coherence

Coherence is an open-source, Git-native drift detector built for repositories where AI coding tools are making frequent changes. The core workflow is a consistency scan: you define relationships between files — docs tied to code modules, tests tied to implementations, generated artifacts tied to source files, API endpoint definitions tied to route handlers — and Coherence checks whether those relationships still hold after each change. It runs locally as a CLI tool, hooks into pre-commit checks or CI pipelines, and the vendor states that deterministic checks do not send code to any external service.

The differentiating design choice is the absence of an LLM in the default check path. Where most drift-detection approaches call out to a model to reason about whether two files are still in sync, Coherence uses rule-based checks defined in an ontology file. That makes results identical on every run and removes the per-call cost and latency that would make a CI gate impractical. The docs describe an optional LLM pass for cases where rule-based checks are insufficient, but it is disabled by default.

Coherence fits tightest in repositories with tightly coupled artifact types — a Go or Python service with ADRs, generated protobuf files, endpoint specs, and test coverage that all need to track each other — used by teams running GitHub Copilot, Cursor, or Claude Code at speed. It does not fit well in loosely structured repositories where the relationships between files have not been formally defined, because the tool can only check links you have told it to watch. Coherence also makes no attempt to remediate drift — it flags, stops, and hands the repair back to you.