Exploring AI tooling, model runners, cloud gateways, and local setups
Like many developers, I’ve integrated AI coding assistants into my daily workflow. While tools like GitHub Copilot were fantastic, recent changes prompted me to look into alternatives.
I’ve spent the last few weeks deep-diving into the landscape of AI tooling, exploring model runners, cloud gateways, and local setups. This review summarises what I found so you don't have to do all the legwork.
My Current Setup (The Basics)
Before diving into alternatives, here’s what I’m currently using:
- Anthropic/Claude: I subscribed directly and use the dedicated VS Code extension. It performs extremely well.
- OpenAI: My employer provides a ChatGPT Enterprise account, which I use via the VS Code extension. It’s also very useful.
The leading proprietary models are genuinely remarkable. However, I wanted to see how the open-source and weight-based options stack up.
The Tool Deep Dive
Here is a breakdown of the services and tools I tested:
Ollama (Local Model Running)
Ollama is great for running models locally, but its performance depends heavily on the service layer you use.
The Good:
- Local Power: Running models like Gemma 3/4 locally (on my workstation with an NVIDIA 3080ti) was surprisingly fast, around 60 tokens per second. This is excellent for creating API endpoints for niche, specific tasks.
The Bad:
- Cloud Reliability: When I tested the hosted cloud models via Ollama, the performance was very poor and unreliable. This made it unusable for demanding, complex agentic coding tasks.
- Warning: Be aware that the hosted service quality seems inconsistent, particularly for heavy usage.
Bottom Line: Fantastic for reliable, contained local tasks. Proceed with caution for demanding cloud or agentic workloads.
OpenRouter (The Universal Gateway)
OpenRouter acts as a unified interface or gateway for accessing many different Large Language Models (LLMs).
The Good:
- Reliability: I found this service much more reliable than Ollama’s hosted offerings.
- Pricing: It uses a prepaid credit system, which is clearer than pay-as-you-go models. You know exactly how much each model costs per million tokens.
- Integration: It integrates seamlessly with tools like the Roo Code VS Code extension and the Zed Editor.
Bottom Line: A reliable, practical option for accessing various powerful models when local resources aren't enough.
Zed Editor (The Blazing Fast Editor)
I wasn't specifically looking for an editor, but I found Zed due to its built-in Agent.
The Good:
- Speed: It is incredibly fast. The editor launches instantly, and the responsiveness of all actions is immediate.
- Agentic Workflow: For coding sessions where speed and responsiveness are key, it works very well, especially when paired with Ollama or OpenRouter.
Things to Note:
- I still use VS Code daily, but Zed is clearly gaining traction and is worth testing for its speed benefits.
OpenCode (Agentic Workflow Helper)
OpenCode is an agentic coding tool designed to work with local and cloud setups.
My Initial Experience:
- When running OpenCode in isolation, I found it lacking. As a coder, I need to see the actual code being generated, and the initial experience wasn't seamless.
My Improved Experience:
- However, when I launched it inside the Zed editor's built-in terminal, its usefulness became much more apparent. Integrating it within a robust editor environment makes a real difference.
Bottom Line: It works best when integrated within a powerful, fast environment like Zed rather than being used stand-alone.
Summary and Recommendations
If you are looking to integrate AI into your coding process, here is a quick guide.
| Scenario | Best Tool/Approach | Notes |
|---|---|---|
| Simple, Niche, Private Tasks | Ollama (Local Setup) | Excellent for running small models locally without cloud costs or latency concerns. |
| Complex, High-Demand Work | OpenRouter + Preferred Model | Use OpenRouter for reliable access to top-tier hosted models without worrying about service stability. |
| Best Overall Workflow | Zed Editor + OpenRouter + OpenCode | The fastest and most integrated experience I found. Pairing the speed of Zed with the reliability of OpenRouter is key. |
| Daily Professional Coding | Stick with VS Code + Paid Assistants (Anthropic/OpenAI) | For maximum familiarity and immediate productivity gains, nothing beats the polished experience of your enterprise-level subscription tools. |
In short: Treat Ollama as a powerful local resource for small jobs, and treat OpenRouter as your reliable gateway for professional, demanding work.