01 / CONTINUITY
01 / CONTINUITY
Keep useful work close.
Pick up where you left off. The chat’s Python session keeps your data and helpers while it runs, and files and history are saved separately.
ABOUT / WHY VARIANT-1
Make room
for the work.
Real work rarely ends with the first answer. VARIANT-1 is built for everything that comes after: a place where a model can run code, keep its working data alive, and improve its tools as the task grows.
OPEN SOURCE SINCE 16 SEP 2026 READ THE LAB LOG
SPECIMEN / V—1
VARIANT-1 at a glance
- Open source
- MIT
- Free to use, read, and change.
- Platforms
- 3
- Windows, macOS on Apple silicon, and Linux.
- Model connections
- 40+
- Built in, plus any OpenAI-compatible endpoint.
- Accounts to create
- 0
- No VARIANT sign-up. You bring the model.
- Live Python session per chat
- 1
- Your data and helpers stay ready for the next message.
01 / PURPOSE
Good work
takes more than a turn.
You compare a set of files, spot something odd, and ask a different question. You build a report, open it, and notice something that needs fixing. Every step leans on the one before.
VARIANT-1 keeps those steps together. Each chat’s live Python session holds on to the data and helpers built so far, and its tools reach your files, the web, and (on Windows) your other apps.
02 / ADAPTATION
02 / ADAPTATION
Make the method visible.
Let the model combine tools and refine its approach. Read the code, check the output, and use what you see to steer the next step.
03 / CHOICE
03 / CHOICE
Choose for the task.
Pick a local model or a cloud provider for each chat, weighing what it can do, what it costs, and what it needs.
02 / WHAT IT’S MADE OF
Four parts,
one workspace.
VARIANT-1 is a desktop app with a Python engine inside. The model you choose decides what to do next, and the engine carries it out on your computer.
LIVING MATTER / STUDY 01
Keep what helps.
Change what needs to change.A DESIGN PRINCIPLE / VARIANT-1
03 / APPROACH
Build around
what helps the work.
Models differ in how they plan, use tools, and recover from mistakes. VARIANT-1 gives every model the same well-equipped workspace, and the guidance it gives them changes as models improve.
The aim is practical: less time rebuilding what was already done, more time moving forward. That means testing real workflows, studying the failures, and fixing whatever gets in the way.
Progress has to show up in the result: a file that opens, a sum that checks out, a change you can see in the app.
04 / CURRENT DIRECTION
A workspace that keeps working.
Your choice of model.
The public preview is out for anyone who wants to put AI to work on code, research, files, and everyday desktop tasks. It’s free, open source under the MIT License, and available for Windows, macOS on Apple silicon, and Linux.
Bring a local model, an API key, or a plan you already have. There’s no VARIANT account or subscription; your provider’s own prices still apply. The app now has long-running goals, automations, and a Review pane for every change, and it keeps evolving in the open: follow along in the lab log.
Download the previewOpen all the
way down.
The source is public under the MIT License. Read the code, report a bug, suggest a change, or run the app straight from source.
git clone https://github.com/Nocturn3529/VARIANT-1.git cd VARIANT-1 npm ci npm run setup:backend npm start
NEEDS PYTHON 3.13, NODE.JS, AND GIT. SETUP GUIDE