Back to recipes
IDE IntegrationBeginner20 minVerified 45 days ago

Use Cline with a Local Ollama Model

Run the Cline coding agent inside VS Code, sending every prompt to a local model so your code never leaves your machine.

clineollamavscodelocal-llmbyokcoding-agent
The promise

Cline is the most popular open-source coding agent for VS Code. It can read files, propose edits, run terminal commands, and manage whole tasks from a chat panel. By default it sends your code and prompts to cloud providers like Claude or GPT-4. That is fast and capable — but it is also expensive, and it sends your intellectual property off-device.

This recipe wires Cline to a local Ollama model. The result: a coding agent inside your IDE that costs nothing per token, works offline, and keeps every line of code on your machine.

What you'll get
  • Cline installed in VS Code
  • Ollama running locally with at least one coding-capable model
  • Cline configured to use Ollama as its provider
  • A verified edit loop on a real file
Prerequisites
  • VS Code installed
  • Ollama running locally with at least one model pulled
  • Cline extension installed from the VS Code marketplace
  • Basic comfort with diffs and terminal output
Sanity checks
Check Command / Action
Ollama running curl http://localhost:11434/api/tags
Model pulled ollama list
Cline installed Look for the Cline icon in the VS Code sidebar
Model selected Open Cline panel, check provider dropdown
GPU active nvidia-smi or ollama ps
Troubleshooting
Symptom Fix
Model list empty in Cline Confirm Ollama is on http://localhost:11434 and at least one model is pulled.
Cline says "provider error" Check that the model name matches exactly, including the tag (qwen3.5:9b, not qwen3.5).
Slow generation Use a smaller model, enable GPU offload, or reduce Cline's max context.
Bad diffs Ask Cline to show the proposed change before applying. Local models are more error-prone on multi-file edits.
Out of memory Stop other Ollama models (ollama stop <model>) or pull a smaller quantized model.
Next step

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

Steps

1

Two models cover most local coding work:

ollama pull qwen3.5:9b
ollama pull qwen2.5-coder:14b

Verify they are available:

ollama list

If you are short on VRAM, qwen3.5:9b is the safer starter. If you have 16 GB+ VRAM, qwen2.5-coder:14b gives noticeably better code reasoning.

2
  1. Open VS Code.
  2. Go to Extensions (Ctrl+Shift+X or Cmd+Shift+X).
  3. Search "Cline".
  4. Click Install on the extension by Cline (formerly Cline Bot).

Cline appears as a new panel in the sidebar, usually on the left.

3
  1. Open the Cline panel.
  2. Click the model selector at the top.
  3. Choose Ollama as the provider.
  4. Select one of the models you pulled, such as qwen3.5:9b.

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 is http://localhost:11434.

  3. Restart the Cline panel if needed.

4

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

"Add a JSDoc comment to the current function."

Cline should:

  1. Read the active file.
  2. Propose an edit.
  3. Show a diff in the chat.
  4. Wait for your approval.
  5. Apply the change when you confirm.

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

5
Task Suggested model
Quick chat / planning qwen3.5:9b
Code generation / refactoring qwen2.5-coder:14b
Reasoning / debugging qwq:32b or minimax-m3:cloud
High-speed classification qwen3.5:9b or llama3.2:3b

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

6

Local models are cheaper and more private, but they are not frontier models. Adjust how you use them:

  • Break large tasks into small, explicit prompts. "Refactor this function" works better than "fix the whole codebase."
  • Use Cline's diff view carefully. Local models can produce slightly malformed edits. Always review before applying.
  • Watch VRAM usage. Running a 14B model with a large context window can exhaust GPU memory. Use ollama ps to monitor.
  • Combine with cloud for hard problems. Keep Cline on Ollama for routine work, then switch to Claude or GPT-4 for the one task that needs maximum reasoning.

Recipe verified Mon Jun 15 2026 00:00:00 GMT+0000 (Coordinated Universal Time). Commands are tested but your environment may differ.

Browse related services