# Local LLM Coding Agent: Aider + Ollama on Windows (No Cloud, No Token Limit, Full Privacy)

I just tried running Aider + Ollama locally on Windows (with Intel GPU) and it is promising! It edits files directly in my repo, auto-commits changes, handles multi-file refactors, adds features, fixes bugs, generates tests... basically anything I need in my .NET codebase.

Zero cloud dependency. Unlimited context. 100% private. $0 ongoing cost.

Here's the streamlined setup that got me productive fast. I used it to create unit tests.

**1. (One-time) Prepare your xUnit test project (if using for tests)**

```
dotnet new xunit -o YourProject.Tests
cd YourProject.Tests
dotnet add package Moq --version 4.20.70
dotnet add package Microsoft.AspNetCore.Mvc.Testing --version 8.0.0 # optional, great for API
dotnet add reference ../YourProject
dotnet sln add YourProject.Tests
dotnet restore
```

**2. Install Aider**

```
python -m pip install --user aider-install
aider-install
```

Windows: Add to User PATH (System → Environment Variables):

```
C:\Users\YourUsername\AppData\Roaming\Python\Python313\Scripts
C:\Users\YourUsername\.local\bin
```

Restart terminals / VS Code.

**3. Grab a strong coding model small enough to run on your GPU**

```
ollama pull qwen2.5-coder:14b
```

The coder variant excels at precise C# syntax, reasoning, and agent-like editing. Around 15s responses on decent hardware, perfect balance.

**4. Launch Aider with your files (from solution root)**

For auto-testing + fixing (great for unit tests):

```
aider ^
  "C:\Repo\YourProject\Api\Services\AuthorizationService.cs" ^
  "C:\Repo\YourProject\Api.Tests\Services\AuthorizationServiceTests.cs" ^
  --model ollama/qwen2.5-coder:14b ^
  --test-cmd "dotnet test YourProject.Tests" ^
  --auto-test
```

Or just for general coding (no auto-test):

```
aider --model ollama/qwen2.5-coder:14b "C:\Full\Path\To\YourFile.cs"
```

**Windows tips**

- Full absolute paths in quotes (spaces break things)
- Use ^ for line continuation in cmd/PowerShell

**What you can do with it (beyond tests!)**

`/add File1.cs File2.cs` → give context

Prompt examples:

- "Refactor AuthorizationService to use proper DI and remove statics"
- "Add a new endpoint for password reset with validation and email"
- "Fix the null ref bug in this method, here's the stack trace"
- "Generate CRUD controller for Product entity + repository pattern"

Aider edits files, shows diffs, commits to Git (with messages), and iterates automatically if you use --auto-test or ask it to fix failures.

**The real wins**

- No API keys / data ever leaves your machine
- Unlimited context, throw your whole repo at it
- Works offline, no subscriptions
- Auto-commits keep history clean
- Feels like a real pair programmer for .NET/C# work

BTW I used IPEX-LLM Ollama for my Intel GPU.

Quick Update: Claude Code runs locally via latest Ollama (free/offline). Same models as Aider. Launch: `ollama launch claude --model qwen2.5-coder:14b`

---

*Co-authored with Claude.*

If this was useful, [you can buy me a coffee](https://ko-fi.com/p47k0).

New articles: [follow via RSS](https://blog.koorevaar.com/feed.xml).
