A Jarvis-style
desktop assistant.

What does “Jarvis-style” mean on a Windows PC?

Imagine you are halfway through a report and want to find the spreadsheet you used last week. You remember the subject but only part of the filename. With VERIDIC, you can describe the file, ask the assistant to look in a particular folder, and check the result while the report stays open.

Here, “Jarvis-style” describes a way of working. You talk or type to an AI assistant that can use tools on your PC. VERIDIC is an independent Windows desktop project. What it can do in a session depends on the tools you enable, the folders you allow it to use, and the services you connect.

Start with voice or text

Try typing “Find the battery-life spreadsheet in my Documents folder and show me the path.” An exact path is easy to copy from a typed reply. Voice is handy when your hands are occupied. With Windows standby listening enabled, say “Hey Veri,” wait for the voice session to connect, then make your request. Standby needs an available microphone and the supported Windows speech recognizer.

Open the Intelligence Feed to see the tool activity VERIDIC reports. For the spreadsheet request, look for the folder and filenames it found. If several files look plausible, open the one you recognize before asking the assistant to summarize any figures. You can also return to a saved conversation in History to find an earlier request.

Give the assistant a window, file, or project

Screen Copilot lets you select a window, capture it, and ask about something visible. If Windows shows a confusing dialog, try “What does the message in this window say, and which setting does it refer to?” You can mark the part you mean. Capture the window again if the dialog changes, especially before asking for a click.

For repeat work, create a Project Space and attach the relevant files. For example, attach two meeting notes and ask “Which decisions appear in both documents? Show the wording from each.” Keeping the originals in the project makes it easier to check names, dates, and the assistant’s interpretation. For a shareable result, review the text and choose an available export format such as Markdown or PDF.

Try a small coding task in Labs

In the Coding workspace, choose the project folder and describe one observable result. A request like “When the search box is empty, show all entries. Add a test for that case and show me the changes” gives you something specific to check. Choose a folder where you are comfortable reviewing edits.

Coding is a Labs feature. Inspect the diff to see which files changed, then look at the commands, exit codes, and verification results for that run. If a required dependency is missing, a test may never start. Run the check once the dependency is available before relying on the change.

Decide what carries into the next conversation

History helps you revisit a conversation. Memory lets you keep a preference across conversations. Try “Remember that I prefer meeting notes with decisions first,” then open Memory to check the wording. You can search, edit, or remove a saved entry. Local memory is available on the device, with an optional Mem0 connection for cloud memory.

Screen Recall has a separate setting. If enabled, it keeps searchable screen text locally. It starts switched off. Review Privacy & Data and your desktop permissions when deciding which parts of your work the assistant can access.

Choose a setup for your first session

Start with the Windows beta installer and a valid VERIDIC licence. Initial activation and periodic licence checks need internet access. Then choose how you want to run the assistant.

• For local text, set up Ollama and download a model your computer can run. Local voice also needs the speech recognition and synthesis models used by Whisper and Piper.
• For the hosted Gemini voice path, configure your Gemini API key and the connection details for your LiveKit project. This path uses the network and those providers to handle the session.

A good first request is to find a file whose location you already know. Check the returned path, then try the same request by voice. That gives you a simple way to check file access and speech setup before handing the assistant a larger job.