Start with ordinary Python
The model can use Python to process data and connect individual tools into a sequence. Files, browsing, and desktop tools provide ways to carry out the parts of a task beyond the calculation.
A simple helper already makes a repeated operation easier to inspect:
def normalise_id(value):
return value.strip().casefold()
normalise_id(" AB-12 ")
# "ab-12"This helper removes surrounding spaces and standardises letter case. Try it on representative inputs before reusing it on a larger dataset.
Tools for different parts of the task
VARIANT-1 groups its tools into Build, Explore, and Operate. They cover files and code, web and desktop work, and capabilities such as asking you questions or using connected services.
The model-facing action retains the historical name ipython. The runtime itself is VARIANT-1's custom CPython environment with top-level await; it is not a Jupyter/IPython session.
You can describe the task in ordinary language; the model handles the tool calls.
Let tool methods adapt
When mutation authoring is enabled, the model can propose and activate chat-local changes to executable tool methods. These changes adapt the method used for the work, not the underlying model weights.
Mutation authoring starts off for new chats. Ordinary helpers and tool composition can be useful without a managed mutation. Adaptation should be judged by whether it helps produce the intended result.
Turning mutation off
Turning the setting off prevents new proposals and activations. It does not remove methods already active in the chat.
Reset and rollback are separate actions. Saved method code does not preserve every live object it depends on; see optional checkpoints.
Check what the method produces
Test a revised method against the output you need, including changed inputs. A successful first run does not show that it will work in every case.
Mutation does not create an operating-system sandbox around ordinary Python. The data and control guide explains the execution boundary.