Setting up Ollama
Run free models on your own computer, with no account and no key.
On this page
Ollama runs AI models on your own computer. Aldwin finds it on its own, so once Ollama has a model, there is nothing to configure. This guide covers installing Ollama, picking a model that works well with Aldwin, and connecting the two.
What Ollama is#
Ollama is a free app that downloads open models and serves them on your computer at http://localhost:11434. Your prompts and files stay on your machine, and it keeps working without internet once a model is downloaded.
The trade-off is speed and quality: a model that fits on a laptop is smaller than the cloud models, and it answers as fast as your hardware allows.
What you need#
- A computer with at least 8 GB of memory (16 GB or more is better)
- About 5 GB of free disk space for a small model
- Aldwin installed (
uv tool install aldwin)
Install Ollama#
macOS and Linux:
curl -fsSL https://ollama.com/install.sh | shOn a Mac you can also download the app from ollama.com/download.
Download a model#
Aldwin's tools need a model that can call tools. On ollama.com, those models carry a tools tag. A good first pick:
ollama pull qwen3:8bOther tool-capable models that run on most laptops are qwen2.5:7b and llama3.1:8b. Each is about 5 GB.
Check what you have:
ollama listConnect Aldwin#
In the web app#
- Make sure Ollama is running with a model downloaded.
- Run
aldwin web. - If this is your first time, pick Free, on this computer on the Connect a model page, then Look for local models. Choose a model and select Use this model.
- Later, open Settings and choose Models. Ollama shows under On this computer. Pick one of its models under Model in use, or from the message box in any chat.
In the terminal#
- Make sure Ollama is running. On a Mac, open the Ollama app. On Linux it usually runs in the background after install.
- Start Aldwin with
aldwin. - On first launch, Aldwin shows
Looking for local models…and lists what Ollama has. Pick a model and you're done. - Later, type
/model. Ollama appears under Available providers with its model count. Pick it, then pick a model. - You see
✓ Switched to qwen3:8b (Ollama, local).
Each model shows ✓ tools or ✗ chat only, and its context size. If you pick a chat-only model, Aldwin warns you and asks Switch to chat-only model? [y/N]: first.
You can also switch directly with /model ollama/qwen3:8b.
Context size#
Aldwin asks Ollama for up to 32K of context for each model, so longer conversations and files fit. That takes more memory than Ollama's own default. If your computer struggles, pick a smaller model.
Ollama on another computer#
If Ollama runs on another machine on your network, tell Aldwin its address in ~/.aldwin/config.yaml:
providers:
ollama:
base_url: http://192.168.1.20:11434On that machine, Ollama must listen beyond localhost. Set OLLAMA_HOST=0.0.0.0:11434 there and restart Ollama. See the Ollama FAQ for each system.
Try it#
aldwinAsk something that uses a tool:
List the files in this folder.
You should see a list_dir step and then the answer. The status line at the bottom shows your model and Ollama. The first answer can take a little longer while Ollama loads the model.
Privacy notes#
- Prompts, files Aldwin reads, and answers stay on your computer.
- Ollama connects to the internet only to download models.
- Details: Privacy.
Troubleshooting#
| What Aldwin says | Fix |
|---|---|
Ollama (not running) in /model | Start the Ollama app, or run ollama serve in another terminal. |
Ollama is not running at http://localhost:11434. | Same: start Ollama. Check with curl http://localhost:11434/api/tags. |
No models found for Ollama. | Download one with ollama pull qwen3:8b. |
Ollama isn't running. Start it and try again. | Ollama stopped during the chat. Start it again and resend your message. |
The model ... isn't installed in Ollama. ... | The model was removed. Run ollama pull for it, or pick another with /model. |
Warning: ... does not support tools. ... | The model can only chat. Pick one marked ✓ tools. |
| The model answers but never uses tools | Small models sometimes skip tools on long prompts. Try /new, a shorter request, or a bigger model. |
| Answers are very slow | The model is too big for your memory. Pick a smaller one, and close other heavy apps. |