VA Intelligence: AI FAQs

Learn more about VA Intelligence

Common questions

Before the 2026.5 release, Explain Symbol performed a Find References operation, then opened an Explain panel that summarized what the symbol was, how it was used in the context of your project, and highlighted key relationships. The panel was generated entirely on your machine using the Gemma model via Ollama; no code or prompts left your computer.

Explain Symbol with AI was removed in Visual Assist 2026.5. To ask about code now, open VA Agent (Tools → VA Agent), switch to Ask Mode, and name the class, function, or symbol you want explained. You can still run Find References without AI.

Local model: Extensions menu → Visual Assist → Intelligence → install language model → enable VA Intelligence in options.

Claude Code or Codex: install the provider's CLI, sign in with your own account, then enable VA Intelligence in options and choose the provider in the VA Agent window.

No. On first use, Visual Assist prompts acceptance of third-party AI terms. Accepting downloads and installs the local AI runtime (Ollama) and model. Claude Code and Codex are never connected unless you sign in to them yourself.
For the local model, Visual Assist provides a download for Ollama (local AI runtime) and a compatible local Gemma model. Using Claude Code or Codex installs nothing from us; you install the provider's CLI yourself.

Yes. You can disable the AI feature at any time in the Visual Assist Options dialog. You can re-enable it later in the same place.

If you want to completely uninstall VA Intelligence, you can go to the options dialog → VA Intelligence → click on Remove files. This will completely wipe any AI-related feature by Visual Assist.

It depends on the model you choose. With the local model, no. Processing is 100 percent local, and your source code, prompts, and results never leave your computer. With Claude Code or Codex, Visual Assist passes the context for the feature you invoke to that provider through your own account, and your account's data terms with Anthropic or OpenAI apply. VA doesn't send data to Whole Tomato servers, and we don't use your data to train any model.
For the local model, only for the initial download of Ollama and the Gemma model from our server. After the one-time install, all local AI work runs with no external calls. Claude Code and Codex need a connection each time you invoke a feature.

Local model:

  • Execution is local on your machine. Your code never leaves your environment.
  • Visual Assist communicates with Ollama through its local web server interface. Traffic stays on the local host.
  • The runtime and model are downloaded from a Whole Tomato server that we manage and version pin.
  • Gemma is a Google model. While we cannot audit its internals, it runs locally under your control.

Claude Code or Codex:

  • Visual Assist acts as the context layer. It talks to the provider's CLI on your machine, which uses your own account.
  • Requests go to Anthropic or OpenAI under your account's terms. Whole Tomato doesn't proxy, log, or store them.
Your choice. A local Gemma model via Ollama (Visual Assist manages the compatible versions), or Claude Code or Codex through your own subscription.
Open the Options dialog in Visual Assist to enable or disable the AI feature and to review its status.
The local model requires a GPU with 12 GB VRAM minimum and approximately 9 GB of free disk space for the model files. CPU-only is not supported for the local model. Claude Code and Codex run on the provider's hardware and don't need a local GPU.
Select code in the editor, then press Shift + Alt + Q and choose Change Code with AI. For chat, open Tools → VA Agent, docked on the right.
Change Code with AI lets you select a block of code, describe the change you want in plain language, review the proposed diff, and accept or reject it. Nothing is applied to your file until you explicitly accept. It works with the local model, Claude Code, or Codex. Pick one from the selector in the Change Code with AI prompt.
Optimizations, renaming for clarity, small refactors, adding error handling, pattern conversions — the long tail of small-but-tedious edits that pile up. Generation-from-scratch works but the experience is rougher; the feature is optimized for modifying existing selections.
Only if you choose a cloud provider. With the local model, Change Code with AI processes entirely on your machine, and your source code, prompts, and diffs never leave it. With Claude Code or Codex, the selection, your prompt, and its context go to that provider through your own account.
Only if you want to use them. VA Intelligence and its local model are accessible with every Visual Assist license. Claude Code and Codex use your own Anthropic or OpenAI account and plan, which you manage with them, and Visual Assist supplies the C++ context. Whole Tomato doesn't sell or bundle model access.
It depends on the feature. Change Code with AI always shows a diff that you accept or reject. VA Agent works like other AI chat tools: in Edit mode it changes files directly, and every edit can be undone. Use Ask mode when you only want answers; it steers the AI toward answering rather than editing.
Contact Sales

Contact Sales

Ask a question or get a quote.

Contact Sales
Purchase Order

Submit a Purchase Order

Get quick turnaround on your orders.

Purchase Order