VibeGrowth

AI policy

Vibe Growth is built to be queried by AI clients, so it should be precise about where AI is actually involved. This page describes what runs today, not what is planned.

Last updated 3 October 2026

The short version

  • Vibe Growth does not train any AI model on your data, and does not supply your data to anyone who does.
  • Vibe Growth does not send your data to a language model on its own initiative. The numbers in the app are computed, not generated.
  • Data reaches an AI model only when you connect your own Claude account (or another AI client) to your workspace and ask it something.
  • When you ask, Claude can read and change your connected advertising and CRM accounts. There is no separate confirmation step inside Vibe Growth.

What is not AI

The figures, comparisons, funnel calculations and anomaly flags in the Vibe Growth web app are produced by deterministic code, meaning explicit rules and SQL against your stored data. There is no model inference behind them.

That is a deliberate choice. An anomaly should be reproducible from its inputs: if the app says spend rose 42% against a seven-day baseline while qualified leads fell, you can check both numbers. A model that merely sounds confident cannot be checked, and would be a poor foundation for a spending decision.

How Claude interacts with Vibe Growth

Vibe Growth runs a Model Context Protocol (MCP) connector. You add it to your own Claude account, where Claude asks you to sign in to Vibe Growth and choose a workspace (only owners and admins can do this). Alternatively you create an access token and paste it into another MCP-capable client such as Claude Code or Cursor.

From then on, the conversation happens in your Claude account, under your agreement with Anthropic. Claude decides what to request from Vibe Growth based on what you ask. Vibe Growth does not choose the model, see the conversation, or control the provider's retention. The data Claude requests is sent to the provider that operates your client.

You control this entirely: no connection or token means no AI access, and revoking one in Settings cuts it off immediately.

What Claude can do through the connector

The connector exposes one tool, a direct passthrough to the API of each platform you have connected (Google Ads, Google Analytics 4, Google Search Console, LinkedIn Ads, Meta Ads, HubSpot). It can read anything the platform authorization you granted covers, and it can modify it: create, edit, pause or delete campaigns, ad groups or ad sets and ads, change budgets and bids, manage Search Console sitemaps, and read and write HubSpot contacts, companies and deals. There is no curated list of actions and no distinction inside Vibe Growth between a read and a change.

Changes are made only in response to your own instructions to Claude. Vibe Growth does not act on its own schedule, and the tool's instructions to Claude say to use a modifying request only when you actually asked for that change. Vibe Growth does not add its own confirmation step. Claude may ask you to confirm in its own interface, but that is Claude's behavior, not ours. You are responsible for the instructions you give, and AI-driven changes can be wrong, so review anything that matters in the platform itself.

The tool's description, which Claude receives, includes the names and identifiers of the accounts connected to your workspace so Claude can address them. Every request through the connector, reads included, is recorded in an audit log held for your workspace; see the privacy policy for what it contains and how long it is kept. The base address of each platform is fixed on our side, so the connector can only reach the platforms you connected.

Skills

Skills are instructions, with optional reference files (text only, no scripts), that you or your teammates write and store in Vibe Growth. They are versioned, and you can download them or invoke them from Claude as slash commands: the connector serves them to your Claude client as prompts when you pick one. Vibe Growth does not write, generate or alter them, and does not run them. When you invoke one, its text is sent to your Claude session like anything else you type, and Claude then acts on it as your instruction, with the same capabilities and responsibilities described above. Do not put passwords, API keys or other secrets in a Skill.

Training

Vibe Growth does not use customer data to train, fine-tune or evaluate any model. We do not train models, and we do not supply your data to anyone who does. How your AI provider treats the conversations in your own Claude account is governed by your agreement with that provider, not by us. Data from Google APIs is additionally covered by the Limited Use commitments in the privacy policy.

If this changes

Server-side AI features are plausible in future — a written summary of a week's performance, for example. If one is introduced, this page will name the provider, say exactly what is sent, and be updated before the feature is switched on. The commitment that customer data is never used for training is not conditional on that.

Questions