Powerful ideas.
Plain-English answers.

Make it comfortable.
Then make it useful.

BRIGGS is a Windows and Android application in active development. This website explains the project; the animated sphere here is not a browser-based version of the assistant.

1. Review the terms. The installed previews present first-launch terms before the assistant starts. Reading the website does not enable the microphone or authorise app actions.
2. Choose your way in. Start with typed input, or enable voice through the app’s permission flow. You can turn the microphone off.
3. Set a comfortable experience. Explore voice, speaking-speed, answer and accessibility preferences. Reduced motion is available if animation is distracting.
4. Prepare local AI if you want it. Model setup and storage requirements depend on the edition. Local tools and model-generated answers are different routes through the application.
5. Try a small, checkable request. Ask for the date, try a supported calculation or explore the available controls before using more involved workflows.

Your words.
Your pace.

The male voice profile uses Kev, with Kevin accepted as an alternative; the female profile uses Lisa. Sir and Ma’am are separate address preferences. A typed alternative remains available.

Speech recognition turns captured sound into text. It is separate from the local AI that generates a response, and separate again from the voice that reads it aloud. Improving one does not automatically make the others perfect.

Why might BRIGGS ask you to repeat something?

The unified recognition workflow can compare recognisers on eligible requests. When transcripts disagree on important details, it may ask again instead of choosing a name, time or action for you. Sometimes the visible transcript looks right even though a second recognition result disagreed.

Patient-listening and optional soft-speech settings can help in supported situations. They do not guarantee accuracy, and boosting quiet speech can also increase background noise. A quieter room and a suitable microphone can still make a difference.

Local intelligence.
Practical shortcuts.

Supported AI models can run on your device. They generate answers from their model knowledge and the context supplied to them; they are not automatically up to date and can confidently make mistakes.

For supported everyday requests, a deterministic tool can calculate an answer directly. Current examples include discounts, bill splitting, serving quantities, weekdays, overnight durations and adding time to a clock.

Try a clear request pattern.

• “Split 100 between 3 people.” — allocates the total to two decimal places; it does not transfer money.
• “Scale 250 from 4 servings to 6 servings.” — returns 375 in the same units; it does not adjust cooking time.
• “Weekday of 10/06/2007.” — reads the date in British day/month/year order.

These are supported examples, not a live demonstration on this page. Device memory, model choice and setup affect the available AI routes and response times.

Shared direction.
Different capabilities.

Choose the edition around what you need to do. The features below are developing workflows, not promises of identical behaviour across devices.

No edition promises control of every app. The assistant cannot override operating-system security restrictions.

Useful assistance.
Meaningful boundaries.

You should know when you are reading an answer, reviewing a suggestion or approving an action. Those are different things.

• Review before acting. Supported sensitive workflows retain confirmation steps. Check recipients, wording, dates and other important details.
• Keep the stop controls. Cancel a response or turn off listening when you need to. Background listening is an explicit choice, not a requirement to use typed input.
• Respect device security. Lock-screen and permission restrictions remain. Hands-free use is not a promise to bypass unlocking for private or sensitive actions.
• Treat external content carefully. Pasted notes, documents and generated answers are information to review—not automatic authority to run commands.

The Windows Hermes integration is deliberately bounded. It is not an unrestricted agent with access to your terminal, files, accounts or messages.

A promising preview.
Not a finished promise.

Automated checks, synthetic speech fixtures and emulator tests help catch problems. They do not replace testing on different real phones, computers, microphones and headsets—or feedback from people with different access needs.

BRIGGS is not an emergency service, medical device or substitute for professional judgement. Check important information, and do not rely on it as your only way to perform a critical task.

The next priorities are more dependable recognition, accessible everyday journeys and broader performance testing. There is no universal accuracy percentage or guaranteed response time.