CLEANUP.md

Cleaning up

None of this is required. If you plan to keep experimenting, keep everything — the model is the slow part to get back, and uv run agentgraph doctor will still pass tomorrow.

But this workshop pulled a multi-gigabyte model onto your machine, possibly installed Ollama and uv, and may have written a variable into your shell profile. If you want your laptop back exactly as it was, here is every piece of it.

The model is the Thinking checkpoint: a separate 8 GB download from the Instruct one agentfix-langchain uses, sharing no blobs with it. If you worked through both editions, both are on disk and each has to be removed by name.

On Option 3 (Colab) there is nothing to clean on this machine. The model, the Ollama install and the clone all lived in the Colab runtime, which is discarded when the session ends. Delete the notebook copy from your Drive if Colab saved one there, and you are done.

What is on your machine, and what put it there

What	Created by	Where it lives	Size
agentgraph-mellum2-thinking + its base model	Option 1	Ollama's model store	~8 GB
qwen3:1.7b	Option 2	Ollama's model store	~1.4 GB
MELLUM_MODEL in your shell profile	Option 2, only if you put it there	~/.zshrc, ~/.bashrc, or the Windows user environment	—
.venv and uv.lock-resolved packages	uv sync	inside the clone	a few hundred MB
uv's package cache	uv sync	~/.cache/uv (%LOCALAPPDATA%\uv on Windows)	up to a few hundred MB
agentgraph-sandbox Docker image	only if you built it	Docker	a few hundred MB
Ollama	only if you installed it for this	per OS, see below	a few hundred MB
uv, and a uv-installed Python 3.12	only if you installed them for this	~/.local/bin, ~/.local/share/uv	~50 MB
Run results	agentgraph solve / eval	results/ in the clone	KBs

Nothing was installed into your system Python: the environment lives inside the clone, so deleting the folder takes the packages with it.

Nothing in the repository was generated. Modelfile, Dockerfile.sandbox and results/precomputed/ all ship with it — there is no setup script here that wrote files into the repo root, so there is nothing to delete except the clone itself.

Open your operating system below. Each block is the complete sequence in the reverse of the order setup did it; skip the steps that do not apply to you.

macOS

1. Remove the model — undoes ollama pull + ollama create. This is almost all of the disk space.

ollama list                                                             # see what you have

# Option 1
ollama rm agentgraph-mellum2-thinking hf.co/JetBrains/Mellum2-12B-A2.5B-Thinking-GGUF-Q4_K_M

# Option 2
ollama rm qwen3:1.7b

Remove the derived model and the base model it was built from. The base model is the multi-gigabyte download; deleting only agentgraph-mellum2-thinking frees almost nothing.

2. Remove the model variable — only if you set MELLUM_MODEL for Option 2.

If you put it in ~/.zshrc (or ~/.bashrc, or ~/.config/fish/config.fish), delete that line. Then, for the terminal you are standing in right now:

unset MELLUM_MODEL

3. Remove Ollama itself — optional, and only if you installed it for this workshop. Do the model first: uninstalling does not reliably take the model store with it.

# installed with Homebrew
brew services stop ollama
brew uninstall ollama

If you installed the app instead: quit Ollama from the menu bar, then drag Ollama.app from Applications to the Trash.

Either way, the leftovers neither route removes:

rm -rf ~/.ollama
sudo rm -f /usr/local/bin/ollama

If you set the loaded-model cap while juggling two editions, undo that too:

launchctl unsetenv OLLAMA_MAX_LOADED_MODELS

4. Remove the Docker sandbox image — only if you built it.

docker rmi agentgraph-sandbox

5. Remove uv and its cache — only if you have no other use for it. A brew-installed uv is a normal package.

uv cache clean
uv python uninstall 3.12                                       # only a uv-installed one
brew uninstall uv                                              # if you installed it that way
rm -rf ~/.local/share/uv ~/.local/bin/uv ~/.local/bin/uvx      # if you used the install script

6. Delete the clone

rm -rf ~/agentfix-react          # wherever you cloned it

That takes .venv, results/ and everything you wrote with it. Copy out any solution you want to keep first.

7. Check it worked

ollama list                          # no agentgraph-… entry
echo "$MELLUM_MODEL"                 # empty

Linux, WSL2 and ChromeOS

1. Remove the model — undoes ollama pull + ollama create. This is almost all of the disk space.

ollama list                                                             # see what you have

# Option 1
ollama rm agentgraph-mellum2-thinking hf.co/JetBrains/Mellum2-12B-A2.5B-Thinking-GGUF-Q4_K_M

# Option 2
ollama rm qwen3:1.7b

Remove the derived model and the base model it was built from — the base model is the multi-gigabyte download.

2. Remove the model variable — only if you set MELLUM_MODEL for Option 2. Delete the line from ~/.bashrc (or ~/.zshrc, or ~/.config/fish/config.fish), then:

unset MELLUM_MODEL

3. Remove Ollama itself — optional, and only if you installed it for this workshop. Do the model first. The install script registers a systemd service, so stop that before deleting the binary:

sudo systemctl stop ollama
sudo systemctl disable ollama
sudo rm -f /etc/systemd/system/ollama.service
sudo systemctl daemon-reload

sudo rm -f "$(command -v ollama)"
rm -rf ~/.ollama

If you added OLLAMA_MAX_LOADED_MODELS=1 with systemctl edit ollama, that drop-in went with the service file above. Inside WSL2 without systemd you started the server with ollama serve, so there is no service to remove — stop that process, then run the last two commands.

The installer also creates a dedicated ollama user with its own model store. Only if nothing else on the machine uses them:

sudo rm -rf /usr/share/ollama
sudo userdel ollama
sudo groupdel ollama

4. Remove the Docker sandbox image — only if you built it.

docker rmi agentgraph-sandbox

5. Remove uv and its cache — only if you have no other use for it. An apt/dnf/pacman uv is a normal package: uninstall it with the same tool.

uv cache clean
uv python uninstall 3.12                                       # only a uv-installed one
rm -rf ~/.local/share/uv ~/.local/bin/uv ~/.local/bin/uvx      # if you used the install script

6. Delete the clone

rm -rf ~/agentfix-react          # wherever you cloned it

That takes .venv, results/ and everything you wrote with it.

7. Check it worked

ollama list                          # no agentgraph-… entry
echo "$MELLUM_MODEL"                 # empty

Inside WSL2, two extra things live on the Windows side. Undo the memory setting if you raised it — edit %UserProfile%\.wslconfig in PowerShell, remove the lines you added, then wsl --shutdown:

[wsl2]
memory=16GB

Or remove the whole Ubuntu installation, which takes the model, Ollama and the clone with it in one command. Only if you installed Ubuntu for this and have nothing else in it:

wsl --unregister Ubuntu

That is irreversible and deletes everything in that Linux filesystem. Copy out anything you want to keep first.

Windows — native PowerShell

1. Remove the model — undoes ollama pull + ollama create. This is almost all of the disk space.

ollama list                                                             # see what you have

# Option 1
ollama rm agentgraph-mellum2-thinking hf.co/JetBrains/Mellum2-12B-A2.5B-Thinking-GGUF-Q4_K_M

# Option 2
ollama rm qwen3:1.7b

Remove the derived model and the base model it was built from.

2. Remove the model variable — only if you set MELLUM_MODEL persistently with setx or the Environment Variables dialog.

reg delete HKCU\Environment /F /V MELLUM_MODEL

Use reg delete, not setx MELLUM_MODEL "" — setx with an empty value leaves the variable present but empty, which reads as a model named "" and fails in a much more confusing way than a missing variable does.

By hand instead: Settings → System → About → Advanced system settings → Environment variables, then delete it from the top (user) list.

Either way, already-running programs keep the old value until they restart. For the PowerShell window you are in:

Remove-Item Env:\MELLUM_MODEL -ErrorAction Ignore

3. Remove Ollama itself — optional, and only if you installed it for this workshop. Do the model first: the uninstaller does not take the model store with it.

Settings → Apps → Installed apps → Ollama → Uninstall, or from a terminal if you installed it with winget:

winget uninstall -e --id Ollama.Ollama

Then the directories it leaves behind:

Remove-Item -Recurse -Force "$env:USERPROFILE\.ollama"
Remove-Item -Recurse -Force "$env:LOCALAPPDATA\Ollama"

4. Remove the Docker sandbox image — only if you built it.

docker rmi agentgraph-sandbox

5. Remove uv and its cache — only if you have no other use for it.

uv cache clean
uv python uninstall 3.12                                       # only a uv-installed one
Remove-Item -Recurse -Force "$env:USERPROFILE\.local\bin\uv.exe", "$env:LOCALAPPDATA\uv"

6. Delete the clone

Remove-Item -Recurse -Force $HOME\agentfix-react     # wherever you cloned it

That takes .venv, results/ and everything you wrote with it.

7. Check it worked

ollama list                          # no agentgraph-… entry
echo "$env:MELLUM_MODEL"             # empty

Afterwards

uv run agentgraph doctor will fail once you have done any of this, which is the point — you removed what it checks for. Everything you built still works, and the exercise test still passes, because it runs against a scripted fake model and never needed a model at all.