vexp

Vexp is a local-first context engine that builds a dependency graph of your codebase and serves AI coding agents only the code they need, cutting token usage by up to 70% with zero network calls.

What is Vexp?

Vexp is a local-first context engine for AI coding agents, built by founder Nicola Alessi's small bootstrapped team in Milan (launched February 2026). It parses a codebase with tree-sitter into a dependency graph of functions, classes, and types, then serves an AI agent only the code relevant to each task — pivots in full, adjacent code skeletonized — cutting token usage by up to 70% with no network calls and no code leaving the machine. It integrates with the major agents (Claude Code, Cursor, Codex, Windsurf, Copilot, Zed) through a single MCP server, auto-configuring itself from the VS Code extension or CLI. Beyond context, it adds session memory and a verify_done tool that mechanically checks completed work against the graph. It reached 10,000+ installs within three months of launching with zero marketing, and launched on AppSumo with a 4.56/5 rating from 25 reviews.

Key Features

• *Local dependency graph** — parses your codebase with tree-sitter into a graph of functions, classes, and types as nodes with calls/imports as edges, stored in local SQLite
• *Context capsules** — returns pivot files in full while collapsing everything else to signatures and return types, bounded to your token budget
• *vep-dc local LLM** — an optional on-device model that preprocesses prompts with intent detection and re-ranking before they reach the cloud agent
• *11 specialized MCP tools** — run_pipeline plus impact analysis, logic-flow tracing, skeletons, index health, and workspace setup
• *verify_done mechanical checks** — verifies parse errors, dangling imports, and untouched dependents with file:line evidence after the agent declares work done
• *Session memory** — remembers what your agent explored and decided, flagging stale context when code changes
• *Multi-repo workspaces** — index several repos as one unified workspace for monorepos and microservices
• *Private by default** — zero network calls, secrets filtered from context, and a git manifest sync for team collaboration

Who is it for?

• Solo developers and teams using Claude Code, Codex, Cursor, or Windsurf who burn tokens re-reading their codebase
• Anyone with a large or complex codebase who wants agents to work from precise context
• Developers on Apple Silicon, Intel Mac, Windows, or Linux who care about code privacy
• Teams in environments with strict data rules that forbid sending source code to third parties

Who is it NOT for?

• Non-developers — this is a developer tool requiring terminal or VS Code use
• Salesforce developers — Apex is not supported yet
• Anyone expecting a visual/graphical product — it is CLI/VS Code-first
• Multi-machine users who do not want per-device licensing costs
• Tiny projects under ~2,000 nodes — that is the free tier