CodeBoundary
A zero-trust sandbox framework that limits which files and commit history an AI coding model can see per session, logging every prompt, response, and data transfer.
CodeBoundary is a data-boundary architecture for teams that use frontier or local AI coding models but need to control exactly what those models can see. It defines data-boundary contracts, declarative rules stating which files, symbols, and commit history a given model session may inspect, and spins up an isolated context bubble for each AI invocation from an IDE or CI pipeline that enforces those contracts.
Within that bubble, the model only sees explicitly permitted files, its output is written to a staging area rather than directly to the working tree until a human approves it, and every prompt-response pair is recorded in an audit trail. It is framed as a governance layer that sits between source code and any AI model, whether hosted externally or run locally, rather than another AI-assisted coding tool itself.
It targets engineering teams that need auditable, bounded AI code assistance rather than open access to a repository.