The Windows workspace, model provider and local code execution have different boundaries. This page summarises behaviour documented in the 1.1.1 user guide; it is not a claim of an independent security audit.
Studio extracts readable content and reports import limits for supported formats, including DOCX and PDF. Adding a file alone does not start a model request.
2 · Choose scope
Review the documents attached to the task
Document selections in chat show what is in scope. Check processing permission and provider terms for unpublished or confidential material.
3 · Send
Selected context may reach the model connection
With a cloud model, selected content and conversation context may be sent to that provider. Choosing a local model does not automatically disable separately requested web operations.
4 · Review
Inspect outputs and records on your device
Check task, analysis and file outputs. A records export may contain research content; inspect the whole export before sharing it.
Local records, API keys and code permission are distinct.
Records and original file
The Windows version keeps work records on the device. DOCX revision creates a new working copy without replacing the imported original and checks the source hash. PDF can be read for review; direct PDF editing is not this workflow.
Keys and model access
An API key stays in memory by default while the app is open. Optional remembering uses Windows protection for the current user. Provider subscription, API account and local model are different connection paths with provider-dependent eligibility.
Local code execution
The project-level local analysis permission is off by default. Once enabled, generated code may run with your Windows account’s file and network permissions. Project scope is not an operating-system sandbox; inspect code and inputs before execution.
The app is free. Model usage depends on your connection.
Studio includes no model subscription or API credit. You may use an eligible provider account, optional API connection or configured local model. Separate agent requests may consume the same account allowance. A failed subscription request does not automatically switch to a paid API.
Calculate with your own tokens and rates
Add tokens across all agent requests. Enter rates from your provider’s current pricing page. Caching, tool calls, other models, tax and subscription terms are outside this simple estimate. The values you enter are not sent to this site’s server.
Create a key in your provider’s official account. Enter it only in Studio’s connection screen; do not put it in messages or screenshots.
Check usage and spending controls in your provider account. An alert is not a cap; OpenAI’s enforced hard spend limit is a separate setting and enforcement may not be instantaneous.
Run Studio’s short connection test yourself. It can make a real request and consume allowance or API credit; check which model answered and when.