You’ve tried every AI coding assistant. Copilot, Cursor, Codeium—they all feel like strapping a jet engine to a bicycle. They’re slow, bloated, and demand your entire GPU just to suggest a for-loop. You’re not alone. Every developer I know has the same silent frustration: Why does this have to be so heavy?
Then I stumbled on something ridiculous. A project called DLLM. It’s a coding agent built directly on llama.cpp, written in the D programming language. Yes, D. The language everyone forgot about except the people who actually use it. And it’s the leanest, fastest AI assistant I’ve seen in years.
Here’s the thing nobody tells you: the bloat in AI assistants is not inevitable. It’s a choice. Most projects pile on dependencies, abstractions, and Python’s runtime overhead until the tool becomes its own problem. DLLM chose the opposite. It goes straight to llama.cpp, skipping the Python middleman, and uses D—a systems language that compiles to native code with zero runtime overhead. The result? A single binary that starts instantly, uses almost no memory, and does exactly what you need: autocomplete, inline suggestions, and code generation—without the circus.
I tested it on a 10-year-old laptop. It worked. No GPU, no cloud API, no waiting. The developer who built it, Danny Arends, didn’t try to reinvent the world. He just wanted a tool that fit into his Unix workflow. That’s the whole story. And it’s beautiful.
But here’s the twist that will make you rethink everything you know about AI tools: The choice of D is not a quirk—it’s a philosophy. D gives you C-level performance with Python-like expressiveness, but without the ecosystem bloat. It’s the language equivalent of a Swiss Army knife that doesn’t come with a cargo trailer. By using D, DLLM avoids the two biggest problems in AI tooling: interpreter overhead and dependency hell. You don’t need to install conda, pip, or a virtual environment. You just run the binary. That’s it.
If you’re a developer who’s tired of the ‘AI assistant that eats your lunch’ trend, DLLM is a breath of fresh air. It’s not trying to be everything to everyone. It’s a tool for people who value speed, simplicity, and control. The comments on the repo already reflect this: ‘Long live the D programming language!’ and ‘next step is a protocol system like ssh that puts the agent virtually inside a system.’ The community gets it.
So here’s my challenge to you: Stop accepting bloat as the price of AI. Try DLLM. It’s open source, it’s minimal, and it might just change how you think about what a coding agent can be. Sometimes the best tool is the one that gets out of your way.
FAQ
Q: Why would anyone use D for an AI tool when Python has all the libraries?
A: Because Python's libraries come with a cost: interpreter overhead, dependency conflicts, and memory bloat. D compiles to native code, runs instantly, and has zero runtime dependencies. If you want a tool that just works without managing a Python environment, D is a pragmatic choice.
Q: Isn't DLLM just a toy project? Can it actually replace Copilot?
A: It depends on what you need. If you want autocomplete, inline suggestions, and code generation without the cloud or GPU, it's production-ready. If you need a full IDE integration with chat and refactoring, you'll miss features. But the point is that it's a viable alternative for those who prioritize speed and simplicity over feature bloat.
Q: Doesn't using a niche language like D limit contributions and long-term support?
A: It's a double-edged sword. D has a small but passionate community, and the project is open source—anyone can contribute. The trade-off is that you avoid the 'bus factor' of Python's ecosystem bloat. If the tool works for you today, it works. Long-term support depends on the project maintainer, just like any other open source tool.