A little guidance.
A lot more independence.

Meet B.R.I.G.G.S: a Windows desktop assistant designed to make everyday computer tasks more hands-free, with you in control.

A project in active development

B.R.I.G.G.S is still being developed. This hub explains the current development build, its intended workflows and its limitations.

How interaction works

Voice requests begin with “Kev”, followed by the assistant’s acknowledgement and your request. Typed requests provide another way to interact. The guides below describe these workflows as they are being developed.

Built to help, not take over

B.R.I.G.G.S brings voice commands, local AI and controlled tools together. Some tasks depend on optional models, internet access or supported applications. It may ask you to clarify a request or approve an action.

Start with “Kev”.

The application is called B.R.I.G.G.S. Its current wake word is Kev. Give it a moment to acknowledge you before making your request.

Getting a clearer response

• Use the microphone closest to you and keep it a consistent distance from your mouth.
• Reduce television or speaker noise. Headphones can help prevent the assistant hearing its own voice.
• Speak at a normal, steady pace. If it mishears a name, check the displayed transcription before repeating the command.

Understanding the microphone indicator

Recognition pauses while the assistant is speaking to reduce feedback. That does not necessarily mean the microphone is disconnected: audio capture can remain enabled. Use the microphone control when you want to turn it off.

Small requests.
Useful results.

Examples from the development build, used after the wake-word acknowledgement. Availability depends on compatible apps, settings and enabled tools.

Open and control

Application launching looks for supported local applications, with website fallbacks where appropriate. Media controls depend on the player exposing an active Windows media session.

Know what is happening

You can check managed background tasks or request cancellation. Cancellation does not undo a completed action or automatically disable a recurring schedule.

Make room for your day

Explore the calendar and reminder tools, your study materials, document search and permitted screen-reading assistance. Check important dates and the results of actions yourself.

Answers you can
trace back.

Bring your own notes, manuals and study materials into a local document library, then explore the passages behind an answer.

Ask about your library

1. Import a supported document through the Documents panel.
2. Ask a focused question using a topic or document name.
3. Follow the returned document citations and check the original context.

New: Evidence brief (Python)

The Documents panel includes an “Evidence brief (Python)” action. Enter a topic to receive selected quotations with document citations.

The bundled worker ranks a limited set of candidate passages using word matching and returns exact excerpts. It does not upload your documents or require an API key. If the worker is unavailable, the app can fall back to its existing keyword search and report that limitation.

Keep expectations realistic

Scanned material may need OCR. Complex tables, layouts and diagrams may not survive text extraction perfectly. Document citations refer to imported content, not necessarily an exact visual position on the original page.

Know what stays local.
Know what goes online.

Privacy guidance should explain what each feature actually does, rather than promise that every part of an assistant is always offline.

Local processing

The document evidence worker runs on your PC without network requests. Local model features also run on your machine when the required model is available. Local processing still uses your computer’s resources and permissions.

Connected features

Internet research, websites and streaming services contact external providers. Their availability and privacy terms are separate from B.R.I.G.G.S. Some integrations may require an account.

Screen, microphone and actions

• Only enable microphone access when you want voice interaction.
• Give explicit permission before using screen-reading assistance. Avoid exposing sensitive information you do not want processed.
• Review recipients, files and proposed changes before approving consequential actions.
• Use document-folder exclusions and memory controls to limit what the assistant retains or uses.

This overview explains development behaviour; it is not a final legal privacy policy.

More ways to get
things done.

The aim is simple: reduce unnecessary clicking and typing, while keeping alternative ways to use the application available.

Choose what works for you

Voice commands can help people who find a mouse or keyboard difficult to use. Typed requests remain useful when speech is tiring, unclear or simply inconvenient. Neither should feel like a second-class option.

A practical starting point

• Position the microphone comfortably and try a short command first.
• Choose a voice and speaking pace that are easy to follow.
• Use permitted screen-reading assistance when visible text is difficult to read, and check important information against the original.
• Keep confirmation steps for messages and changes that matter.

Honest about the limits

Speech recognition can struggle with background noise, different accents and some speech patterns. Screen text extraction can make mistakes. B.R.I.G.G.S is not a medical device, and these features do not establish accessibility certification or suitability for every individual need.

Let’s get you
back on track.

Start with the smallest useful check. Avoid changing several settings at once, so you can tell what helped.

It does not hear me

Check the selected microphone, its Windows input level and desktop-app microphone permissions. Make sure the app’s microphone is enabled. Say “Kev”, wait for the acknowledgement, then give one short request.

It heard the wrong command

Check the transcription. Reduce background noise and repeat the exact app name. Do not confirm an unexpected action just to move past the prompt.

An answer is taking too long

Ask for “Background work status”, or use “Cancel background work” for managed tasks. Model loading and document work can take time. Cancellation requests do not reverse actions already completed.

The evidence worker is unavailable

The app can report a keyword-search fallback when the Python worker cannot run. That fallback is useful, but it is not proof that the Python worker ran successfully. Worker availability and failure reporting remain part of development testing.

Better evidence.
Clearer boundaries.

This internal build connects a focused Python document worker to the existing Windows application, without replacing the desktop interface or its local models.

What changed

• Added a Python-powered evidence brief with exact document quotations and citations.
• Integrated passage selection into relevant document, tutoring and coding-context workflows.
• Added bounded worker execution, cancellation, response validation and a visible fallback route.
• Bundled the Python runtime needed by this worker.

What was checked

The original feature report records 1,230 regression checks and 16 standalone Python tests, alongside targeted interface, boot, voice-lifecycle and local-model checks.

What this internal build does not claim

It is not a newly trained general-purpose AI, universal autonomous coding system or independent fact-checker. B.R.I.G.G.S remains a work in progress, and optional model features have separate runtime requirements.

One assistant.
Purpose-built components.

The desktop shell stays in C#. Specialised AI features can use Python through a controlled bridge, with existing open models supplying language and speech capabilities.

How a document brief is assembled

1. The application selects a bounded set of candidate passages from the local library.
2. A fixed Python worker ranks those candidates and selects exact excerpts.
3. The C# bridge checks the response format, passage references and quotations.
4. The library rechecks document integrity before publishing citations.

Reliability by design

The worker uses a single shared execution slot, a deadline, bounded input and output, and cancellation. It does not accept a model-generated shell command as its program.

Important boundary

An isolated Python invocation is not an operating-system security sandbox. The worker still runs with the Windows user’s privileges. Tool permissions, input validation and explicit approval remain important.