SMF
Back to deployment recipes
IDE Integration

Use Cline with a Local Ollama Model

clineollamavscodelocal-llmbyok

What you'll get

Cline running inside VS Code, sending prompts to a local Ollama model instead of a cloud API. This setup keeps your code and prompts on your machine and removes per-token billing.

Prerequisites

  • VS Code installed
  • Ollama running locally with at least one model pulled
  • Cline extension installed from the VS Code marketplace

Step 1: Pull a local model

ollama pull qwen3.5:9b
ollama pull qwen3-coder:32b

Verify:

ollama list

Step 2: Install Cline

Open VS Code, go to Extensions, search "Cline", and install it. Cline appears as a new panel in the sidebar.

Step 3: Switch Cline to Ollama

Open the Cline panel and click the model selector. Choose Ollama as the provider. Cline will list the models available at http://localhost:11434.

Select qwen3.5:9b or qwen3-coder:32b.

If the model list does not load:

  1. Confirm Ollama is running:
    curl http://localhost:11434/api/tags
  2. Make sure Cline's Ollama host setting points to http://localhost:11434.

Step 4: Verify the loop

Open any code file and ask Cline a small, safe task:

"Add a JSDoc comment to the current function."

Cline should:

  1. Read the file.
  2. Propose an edit.
  3. Show a diff.
  4. Apply it when you approve.

Watch nvidia-smi or ollama ps to confirm the local model is loaded.

Step 5: Pick the right model for each task

Task Suggested model
Quick chat / planning qwen3.5:9b
Code generation qwen3-coder:32b
Reasoning / debugging minimax-m3:cloud or gemma4:12b

Switch models inside Cline's provider dropdown. You do not need to restart VS Code.

Step 6: Keep costs at zero

Because you are using Ollama, Cline's only cost is your local electricity and hardware. There is no API meter running. This makes it ideal for:

  • Iterating on prompts
  • Testing new models
  • Sensitive code you do not want to send to a third party

Trade-offs

  • Local models are usually less capable than frontier cloud models. Complex multi-file refactors may need Claude or a larger cloud model.
  • You manage model downloads, VRAM limits, and quantization yourself.
  • Initial setup is slower than clicking "Use Claude."

Next step

Combine this with the Ollama CUDA recipe for GPU acceleration, or add Aider for terminal-based git-aware edits on the same local model.