Workspace & BillingToolsXpo Code Reference

Where your code goes

Your code goes to the model you chose, per request, so it can answer — and nowhere else. What that means in practice.

Updated September 2, 2026

An AI coding agent has to send code to a model to get an answer. What matters is which model, and what happens to it after.

This page covers the coding agent. For AI tools called over the API, see What happens to what you send an AI tool.

The path

The agent reads the files it needs from your machine.
They are sent, with your prompt, to the ToolsXpo gateway.
The gateway forwards the request to the provider that serves the model you picked.
The answer comes back the same way.

ToolsXpo does not retain the content. It is forwarded to serve the request and not stored. What is kept is metadata — model, token counts, latency — which is what your usage page and your bill are built from.

Which provider receives it

Whichever one serves the model you chose. The picker names the vendor for every model, so you know before you send.

If you use your own key, the request goes to that vendor under your own account and their terms apply to you directly.

Provider retention

Each provider has its own policy for data sent to their API. If a specific provider's terms matter to you, choose a model from a vendor whose terms you accept — the picker's vendor: filter narrows to one.

Prompt caching means a provider may hold a prefix briefly to serve a subsequent request faster. Cache lifetimes are short — minutes — and provider-specific. See Prompt caching.

Controlling what is sent

.toolsxpoignore — the strongest control. Works like .gitignore; matching files are not read and cannot be sent, even if you ask. This is where secrets, credentials and anything you cannot send belong.

Plan mode is not a boundary. It does not edit files, but it reads and sends them exactly as Code mode does.

Scouts read too. They run on a cheaper model but read the same files. If a path must not leave your machine, .toolsxpoignore it rather than relying on which model is running.

Local models

Point a workspace at a locally hosted model and nothing leaves your machine at all. Add it under Dashboard → AI → Models.