Skip to main content

Security and Privacy of Local AI Context

Vibgrate AI Context runs locally, needs no account, and uploads nothing. Learn exactly what stays on your machine and why the local context server is safe for sensitive codebases.

Vibgrate Docs

Vibgrate Help

When you feed an AI assistant context about your code, where that context lives matters. Vibgrate AI Context is built local-first: it runs on your machine, requires no account, and uploads nothing. This article explains the privacy and security posture in detail.

It is for security teams, regulated organizations, and any developer working on sensitive code who needs to know exactly what leaves the machine.

Overview

Vibgrate AI Context — the local, offline MCP server started with vg serve — serves your code map, drift, and version-correct library docs to your AI assistant directly, on your machine. There is no Vibgrate Cloud round-trip involved in serving that context.

What stays local

  • Your source code is never uploaded by the local context server.
  • Your code map is built locally by Vibgrate Graph and served locally.
  • Your drift (DriftScore) is computed and served locally.
  • Locally ingested library docs stay on the machine.

No account is required to run vg serve.

The one network touchpoint

The only outbound step is a one-time download of a small local model for semantic search, plus optional fetches from the hosted Free Dev Docs Library catalog. You can complete the model download in advance:

vg embed

After that, the server runs fully offline. In air-gapped environments, rely on locally ingested docs rather than the hosted catalog.

Local vs hosted, by design

Vibgrate deliberately separates surfaces:

  • Vibgrate AI Context is local and offline (vg serve) — nothing uploaded, no account.
  • Vibgrate Cloud and the hosted Vibgrate MCP are opt-in. Data only reaches the cloud when you explicitly push it, for example with vg push after authenticating via vg login.

If you never push, nothing leaves the machine through Vibgrate.

Why this is safe for sensitive codebases

Because serving context is local, the assistant gets grounded answers without your code traversing a network you do not control. You get the accuracy benefit of grounded context without the exposure of sending source to a third party.

Related

  • Offline and air-gapped AI context
  • What Is Vibgrate AI Context?
  • Starting the local context server with vg serve

Related Commands