Troubleshooting

Most problems come down to one of three things: Ollama isn’t running, the model is too big for your RAM, or Windows is being cautious. Here’s how to clear each.

”No models” / Pumkin can’t reach a model

Pumkin needs Ollama running to do anything. If it reports no models or can’t connect:

Check Ollama is running. Open a terminal and run ollama list. If it doesn’t respond, Ollama isn’t up.
Start it. On Windows, Ollama runs as a service that auto-starts — restarting your machine usually brings it back. You can also relaunch Ollama from the Start menu.
Confirm you’ve pulled a model. ollama list should show at least llama3.2:3b. If it’s empty, run ollama pull llama3.2:3b.
Restart Pumkin after Ollama is confirmed running, so it re-checks the connection.

The model is painfully slow or won’t load

Almost always a memory problem — the model is too large for your RAM.

On 8 GB, use llama3.2:3b. Larger 7–8B models (like qwen3:8b) will swap to disk and become unusable. This isn’t a bug; it’s physics.
Close other memory-hungry apps (browsers with many tabs are the usual culprit).
If you pulled a big model to try it and it’s grinding, switch the agent’s model back to llama3.2:3b in the dropdown.

See Install Ollama & models for the RAM-to-model guide.

”Windows protected your PC” on launch

This SmartScreen warning appears because Pumkin isn’t code-signed yet (a paid certificate is on the roadmap). It’s expected and harmless for the installer you got from your purchase link.

Click More info → Run anyway.

The download link isn’t working

The link is tied to your purchase — use the one from your welcome email.
If the file is gone, the same link re-downloads it.
Still stuck? Email hi@pumkin.app with the address you bought with and I’ll resend.

The desktop shortcut icon looks generic

A cosmetic Windows icon-cache quirk that sometimes shows a stale icon on the shortcut even though the app itself is fine. It doesn’t affect anything. Signing out and back in (or a restart) clears the cache. Safe to ignore.

An agent did something unexpected

Open the run and read the events in order — the tool_call, tool_result, and following model_response usually reveal the cause. Most often it’s a vague system prompt or a tool returning something the model misread. See Runs & threads for how to read a run.

Where the logs are

If you need to dig deeper, Pumkin’s data and logs live under:

%LOCALAPPDATA%\Pumkin\

When you email support about a bug, mentioning what the run showed (or attaching what’s in that folder) gets us to an answer faster.

Still stuck?

Email hi@pumkin.app. One founder, real support — describe what you did, what you expected, and what happened, and I’ll get back to you, usually within a day.