THE PRODUCT / A LIVING WORKSPACE

Give your AI
room to work.

VARIANT-1 gives your model a live Python session, tools for files, the web, and your desktop, and room to keep going. Ask for a result, watch it come together, and carry on in the same chat.

License: MIT, free
Runs on: Windows, macOS, Linux
Model: Local, API key, or a plan
Account: None needed

FIG. 1 / ANATOMY OF A WORKSPACE

Your model does the thinking; the cell around it does the work. Select a part to see what it does.

01 / THE WORKSPACE

From the question
to the working file.

Bring a project, a folder of documents, or a job in another app. The model works through it with code and tools, while the chat, files, browser, terminal, and Review pane keep everything in view.

01 Ask
02 Work in Python
03 Use the tools
04 Review

Explore example workflows

VARIANT-1 with chat history on the left and a conversation in the middle: two requests about a sales file, each answered with a short explanation and a table of totals by region. The Review pane on the right shows the change to summary.py and a new report file, next to the project’s file list. — Sample project. The conversation was set up for this picture; everything around it is the app as it is today.

02 / YOUR LAYOUT

Pull any panel out.
Put it anywhere.

Files, Review, the browser, and the other side panels can each leave the main window and become a window of their own. A chat, the Log monitor, and Overview can open in their own windows too, so the work can spread across one screen or several.

Keep an eye on Overview or the logs while you work in another app, then send everything back into place when you’re done.

Detach: Opens a side panel in its own window. Chats have Detach chat in their menu.
Pop out: Moves Overview or the browser out of the main window. The Log monitor always opens in its own window.
Pin: Keeps a window above your other apps. The Log monitor starts pinned.
Dock: Returns a panel to the main window.

SCREEN RECORDING / PREVIEW, OCTOBER 2026

Files and Review detach, Overview and the Log monitor open beside them, then everything docks back. Recorded in the app with a sample project; the chat was staged.

03 / LIVING STATE

Continue from
what you already have.

Clean a dataset, then ask for a different comparison. The model reuses the data and helpers from earlier steps for as long as the chat’s session runs. Overview → Python shows every live session and what it’s doing.

A restart clears live data. Checkpoints can keep supported values if you turn them on, and the chat’s own history stays searchable through session context.

See what lasts

EXAMPLE / LIVE SESSION

LIVE WORKING STATE

Ready for the follow-up.

recordsalive

cleaned_recordsalive

normalise_id()alive

The data and the helper are still in the session. The next step can use them straight away.

04 / EVOLVING TOOLS

Build a method.
Refine it as you go.

Your model turns repeated steps into Python helpers and combines them with other tools. The code stays readable, so you can see exactly how the work gets done.

Turn on mutation and it can go further: propose a better version of a tool, test it, and switch it on for the chat. Every change is recorded, and any change can be rolled back.

Read about tools and mutation

Mutation is off by default and never retrains the model. Turning it off stops new changes; changes already switched on stay until you roll them back.

METHOD LAB / normalise_id() Mutation

LINEAGEILLUSTRATION, NOT THE APP INTERFACE

05 / GOALS & AUTOMATIONS

Work that
outlives a chat.

Start a Goal and the agent keeps a bigger objective on track: a plan, progress notes, and anchors to pick up from. It can bring in helper agents for parts of the job, and you can see what each one is doing.

Automations run an instruction on a schedule or when something happens on your computer. With the optional messaging gateway, you can even check in from Telegram or Discord.

How goals and automations work

Goals and automations call your model like any chat does, so keep an eye on usage, and pause anything you aren’t watching.

COLONY / A GOAL FINDING ITS ROUTE

Click the plate to add a step to the plan. Helper agents show as white points. Illustration, not the app interface.

06 / OVERSIGHT

See everything
it did.

The Review pane shows file changes as readable cards with line numbers, lets you browse Git history across branches, and you can ask the agent to commit, push, or open a pull request.

Activity marks steps that failed, were stopped, or ended unclear, instead of calling everything a success. Overview shows live sessions, their CPU and memory, and the logs.

You can steer a task while it runs, queue your next message, or stop it. Stopping can’t undo what already happened, so the record matters.

How you stay in control

REVIEW+2 −1

reports/summary.py 1 file

12records = load("sales.csv")
13−ids = [r.id for r in records]
13+ids = [normalise_id(r.id) for r in records]
14+assert len(ids) == len(set(ids))
15total = sum(r.value for r in records)
··· 18 unchanged lines

BRANCH clean-ids → main3 COMMITS

Ask the agent to commit, push, or open a pull request.

ACTIVITYLAST RUN

Read three files
Request failed, retried once
Stopped by you
Outcome unclear: check the folder

OVERVIEW · PYTHON2 LIVE

Chat · sales: CPU
: MEM
Chat · report: CPU
: MEM

07 / MODEL CHOICE

Choose the connection.
Keep the workspace.

Run a model on your own machine, add an API key, or sign in with a plan you already have. Pick the model for each chat, and line up backups in Settings → Model routing.