Tips & Tricks

Set up Cline with Ollama for a free local coding agent in VS Code

2026-09-12 - ABikram Mondal

Why a local agent beats hosted free tiers for daily work

Most free tiers cap completions or chat requests at a few thousand per month. Once those run out you hit a paywall or slower queues. A local setup using Ollama and Cline removes every external quota because the model runs on your machine.

Cline is an open source VS Code extension released under Apache 2.0. It acts as an agent that can read files, edit code, run terminal commands and browse when you approve each step. It connects to any OpenAI compatible endpoint, including the one Ollama exposes at localhost.

Ollama itself stays free with no account required. You download a model once and it serves requests through an OpenAI style API on port 11434. No tokens leave your computer unless you choose a cloud model later.

The real limit is your hardware. A 14 billion parameter model needs roughly 9 to 16 GB of RAM and works on CPU alone, though GPU makes it faster. Larger models demand more memory and can feel sluggish on older laptops common in many Indian households.

Privacy improves because nothing travels to a vendor server. Code stays on your disk during the session. This matters for proprietary work or when internet is unreliable.

Install Ollama on Windows, macOS or Linux

Go to ollama.com and download the installer for your operating system. The package adds a system service so Ollama starts automatically.

After installation open a terminal and run ollama --version. You should see the current release number. If the command is not found, restart your terminal or add the Ollama folder to your PATH.

On Ubuntu or other Debian based systems you can also use the one line curl command listed on the site. It pulls the script and installs the binary plus the service file.

Windows users get a desktop app that runs in the background. macOS users see a menu bar icon. Both expose the same localhost server.

Verify the server responds by running curl http://localhost:11434 in another terminal. You should get a simple JSON reply confirming the service listens.

Download a coding capable model

Run ollama pull qwen2.5:14b in the terminal. The download is about 9 GB and takes time depending on your connection. Smaller 7B models finish faster but give weaker results on longer refactors.

Once the pull completes, test it with ollama run qwen2.5:14b. Type a short prompt like explain a Python decorator and watch the reply stream back. Exit with /bye when done.

You can list available models anytime with ollama list. The command shows size, modified date and the tag you pulled.

If your machine has limited RAM, start with qwen2.5:7b instead. It needs roughly 5 GB and still handles most single file edits and explanations well.

Models stay on disk until you delete them with ollama rm. You can pull multiple and switch between them later in Cline settings.

Add the Cline extension to VS Code

Open VS Code and search the Extensions marketplace for Cline. The official entry shows the Apache 2.0 license and recent updates. Install it.

After installation a new sidebar icon appears. Click it to open the Cline panel. The first launch asks for an API provider and key, but you will point it at Ollama instead.

Restart VS Code once so the extension registers fully. Check the output panel for any startup errors related to the local server.

Cline works alongside other extensions. You can keep Codeium or Copilot free tiers active if you want autocomplete from multiple sources at once.

Point Cline at your local Ollama instance

In the Cline panel open settings and select OpenAI Compatible as the provider. Set the base URL to http://localhost:11434/v1. Leave the API key field empty or type any placeholder string.

Choose the model name exactly as it appears in ollama list, for example qwen2.5:14b. Save the configuration.

Test the connection by asking Cline a simple question in the chat panel. It should reply using the local model without any network call outside your machine.

Enable agent mode in the same settings panel. This lets Cline propose file edits and terminal commands. You approve or reject each action before it runs.

Context length defaults work for most files. Increase it in Ollama settings if you work with very large repositories and have enough RAM.

Daily workflow and the hardware catch

Open a project in VS Code and use the Cline sidebar to describe a task. The agent reads relevant files, suggests changes and waits for approval. Approve, watch the diff apply, then test the result.

Response speed depends on your CPU or GPU. A recent laptop with 16 GB RAM handles 14B models at usable speed for short sessions. Older machines or those with only 8 GB feel noticeably slower on complex refactors.

No monthly reset occurs. You can run as many requests as your hardware and electricity allow. The only recurring cost is power and any future hardware upgrade.

Updates to Ollama or Cline arrive through their normal channels. Check the GitHub releases for Cline and the ollama.com blog for model improvements.

If the local model struggles with a task, you can switch the provider in Cline settings to a paid API for that session only. Most developers stay on local for routine work and use cloud only when needed.

The short version. Cline plus Ollama gives you an unlimited local AI coding agent in VS Code once you install both tools and point the extension at localhost.

Sources

Reported from the sources above on 2026-09-12. Figures are as published at the time of writing. If something here has moved on, the linked source is the one to trust.

From the desk of ABikram Mondal

If you got here because you are actually thinking about having it built properly once, instead of stitching six free tiers together and maintaining them forever, that is the work I do. I build for founders and small teams who want the thing to exist and work, not a deck about it.

Web Development services · everything I do · talk to me