Vexp

Vexp is a local‑first context engine that supplies AI coding agents with only the portions of a codebase they need to process. By building a dependency graph, identifying pivot nodes, and generating a skeleton context, it reduces the amount of code sent to the model, achieving a 65‑70 % token reduction. All indexing, traversal, and capsule creation happen on the user’s machine, with no network calls, accounts, or cloud services involved.

The tool supports roughly thirty programming languages and integrates with twelve AI agents, such as Claude, Cursor, and Copilot, through a set of specialized commands (e.g., run_pipeline, get_context_capsule, search_logic_flow). It works with existing workflows, including git‑native manifests and cross‑repository contexts, and can be invoked from VS Code or the command line via the vexp‑cli package.

Designed for developers who want deterministic, privacy‑preserving assistance, Vexp aims to make AI‑driven coding more efficient by feeding agents a concise, relevant subset of the repository, thereby extending token budgets and reducing unnecessary processing.