Why is it slow or broken? (It's usually not your computer.)

Checked 22 October 2026. US dollars. No affiliate links.

1. 1

   It forgets long documents, or agent tools fail.

   Ollama gives most machines only a 4K memory window by default; agent tools need 64K or more (Ollama docs). Raise the context setting, then run ollama ps and check the CONTEXT column.

2. 2

   It suddenly crawls.

   The model is too big and spilled off your graphics card. ollama ps should say GPU, not CPU. Pick a smaller model or a shorter context.

3. 3

   It won't fit even though I have 16GB.

   Your system and apps use 4–7GB. Plan on about 60% of your memory for the model.

4. 4

   I don't understand the file names.

   Q4_K_M is the good default. Q8 is twice the size. Avoid Q2 and IQ2. GGUF works everywhere; MLX is Mac-only and a bit faster.

5. 5

   It's not as smart as ChatGPT.

   Small local models are for drafts and summaries. For hard work, use a hosted open model through a cheap plan or a router.

6. 6

   I thought Ollama was private.

   Models tagged :cloud run on Ollama's servers, not your computer.

7. 7

   The model isn't in my app.

   Stick to models listed in your app's own catalog.

8. 8

   Agents don't work on my laptop.

   Tool use needs about a 20B+ model and roughly 23GB of graphics memory. Use a hosted model or a $10–20 plan instead, or see agents on local models.