Start here
Use the kit
Rented a GPU? Here's your first hour.
Fig's kit is one small shell script you can read. It gets a fresh GPU doing something useful, then helps you leave with your work. Run these in the terminal of the GPU machine you rented.
Step 1: Install (one line)
In your GPU's terminalmkdir -p ~/.local/bin && curl -fsSL https://raw.githubusercontent.com/fig-the-fox/fig-kit/fb44aac672b0d10844e37192f8cc1e34446e8c66/fig-v0.3.1.sh -o /tmp/fig.sh && echo "e31548732db26797c887858289d8079d814cf21fd09532c6c26ae72383ec93e2 /tmp/fig.sh" | sha256sum -c && install -m 755 /tmp/fig.sh ~/.local/bin/fig && export PATH="$HOME/.local/bin:$PATH"
It checks the file's fingerprint before installing anything. No sudo. If the check fails, nothing is installed.
Step 2: See what you got
fig doctor
Your GPU, its memory, CUDA version and free disk, in a few seconds.
Step 3: Pick a recipe
fig start chat
Asks first, then installs Ollama (its official installer) and downloads an open AI model sized to your GPU. It runs only on that machine, with an API at 127.0.0.1:11434. Then chat with ollama run llama3.1:8b. The first run takes a few minutes.
fig start persona
Gives the chat model from fig start chat a fox personality: Fig, a friendly, honest guide. It installs and downloads nothing, so run fig start chat first. Then chat with ollama run fig-fox.
fig start notebook
Starts a private Jupyter notebook locked to the machine, with a login token. It prints the SSH tunnel line to open it from your laptop.
Step 4: Before your time's up
fig timer 2
Counts down two hours and warns you in your terminal at 15 and 5 minutes left.
fig save
Packs ~/fig-out into one file and prints the command to download it before the rental ends. Cache it before you go.