AI Transparency

G is your assistant inside Gbuild. When you talk or type to G, AI helps carry out your request — answering questions, creating content, working with your files and connected services, and managing tasks. AI also powers features you may not interact with directly, such as transcribing your speech, classifying your intent, organizing project memory, and identifying sensitive documents for the on-device Vault.

In every case, AI acts as a tool under your direction. It does not make decisions on its own about things that matter — it assists, and you decide.

Gbuild runs several AI capabilities directly on your Mac, so your data stays on your device for these tasks:

• Speech-to-text — your voice is transcribed locally using on-device models. The audio does not leave your Mac.
• Text-to-speech — when G speaks to you, the speech is generated on your device.
• Intent classification — understanding what you are asking G to do is processed on-device using a local model, with cloud fallback only when the local model is unavailable or when real-time voice mode requires it.
• Embeddings and project memory — your project knowledge base is indexed using on-device embeddings, so your project context stays local.
• Sensitive-file classification— the Vault's document classifier runs on-device to identify financial, identity, and medical documents without sending them anywhere.
• Local chat — where your hardware supports it, Gbuild can run smaller language models locally for conversational tasks, keeping the entire exchange on your Mac.

On-device processing is preferred wherever the task allows. When a capability runs locally, no data for that task is sent to any server.

Some tasks benefit from larger, more capable models that require cloud infrastructure. When cloud processing is needed, Gbuild routes your request securely through its own gateway service to the AI provider that fulfills it:

• Complex conversations — questions or tasks that need a larger model than what runs locally.
• Image generation — creating images from your descriptions or editing existing images with AI.
• Video generation — creating video content from prompts or reference images.
• Real-time conversational voice — when G speaks back to you in a natural conversation, audio may stream to a cloud provider.

When a request goes to the cloud, only the content needed to fulfill that specific request is sent. Gbuild routes through its own gateway — we do not embed provider API keys in the app or send your requests to providers you have not authorized. We work with leading AI model providers selected for their capabilities and data-protection practices.

We do not use the content of your requests, your conversations with G, your creative work, or your personal data to train our own AI models. Your data is yours — it powers your experience, not ours.

Cloud AI providers process your requests under their own terms. We choose providers with strong data-protection practices and route requests through our own gateway to maintain a consistent security posture.

You stay in control. Gbuild is designed so that:

• You can review and modify. AI-generated responses, content, and actions are presented to you — not silently applied. You can edit, redo, or discard anything G produces.
• Consequential actions require approval. When G is about to do something that changes your data — sending a message, modifying a file, running an automation, committing code — it asks for your confirmation first. Agent-initiated actions are off by default.
• Voice commands are confirmed. Sensitive voice intents are routed through a confirmation gate: Gbuild shows you what it heard and waits for you to approve before proceeding.
• You can reverse. Where possible, actions taken by G can be undone. We design for recoverability.

When Gbuild generates content using AI — text, images, video, or other media — it is presented as AI-generated. We believe in honesty about what AI creates. When AI-generated content could reasonably be mistaken for human-created content, clear disclosure is the right default. We encourage the same transparency from everyone who uses Gbuild to create and share.

AI is a powerful tool, but it is not infallible. Language models can produce inaccurate information, miss context, or generate content that does not match your intent. Image and video generation can produce unexpected results. Intent classification can misinterpret what you said.

We build Gbuild with this reality in mind: confirmation gates, the ability to review before acting, and clear feedback about what G is doing are all designed to keep you informed and in control, even when AI gets it wrong.

The conversation around AI transparency is evolving, and so are we. We follow emerging expectations around responsible AI — including transparency obligations, labeling standards, and accountability frameworks — and we intend to meet or exceed them as they develop. We would rather lead than catch up.

If you have questions about how AI works in Gbuild, or feedback on how we can be more transparent, email us at [email protected]. For privacy-specific questions, see our Privacy Policy or email [email protected].