AI Context Drift

AI context drift is the mismatch between what an AI coding assistant believes about your stack and what your repository actually runs. Assistants are trained on a snapshot of "latest" plus whatever generic documentation they retrieve — but your project runs the specific versions recorded in its lockfile, with its own structure and conventions.

What It Looks Like

The symptoms are familiar: code written against APIs that were deprecated in the version you run, imports from packages you do not use, hallucinated method signatures, and confident edits that misread how the repository is actually wired together. The assistant is not wrong about some codebase — it is wrong about yours.

Closing the Gap

The fix is to feed the assistant version-correct context instead of generic context. Vibgrate AI Context serves library documentation pinned to the exact versions in your lockfile — plus a deterministic map of your real code and your offline drift results — to any AI assistant over local MCP, without your code leaving your machine.

Related Terms

AI context drift is one axis of code drift; it compounds dependency drift, because an assistant writing against wrong versions creates upgrades you did not plan for.