How I built Claude Code’s workflow with a local model

Ethan Wright 09/10/2026

Building Your Own AI Code Assistant: Recreating the Power of Local LLMs

The world of AI coding is rapidly evolving, with powerful models demonstrating incredible capabilities in inspecting code repositories, making edits, executing terminal commands, and verifying the results of those actions. These advanced functions, often exemplified by tools like Claude Code, showcase how large language models are transforming software development.

But the true magic often lies not just in the model itself, but in the workflow built around it. The experience of truly effective AI coding is often defined by the seamless interaction between the model and the tools it uses.

One curious experiment sought to replicate this sophisticated workflow, bringing the power of cutting-edge LLMs closer to the user. The goal was simple: to achieve these powerful coding automation tasks entirely on a local machine, removing the dependency on external services.

To achieve this, the experimenter combined several open-source tools to create a personalized, self-hosted environment. The setup involved integrating the Qwen 3.5 model with LM Studio and the Hermes Agent framework. This combination allowed for a holistic system where the language model could interact with the operating environment and files on its own computer.

While this local setup was a significant step toward achieving a fully private and self-contained solution, it came with its own set of realities. Running the 14B model, while proving highly functional, did not quite match the performance ceiling set by more commercial and larger models.

Despite this difference, the resulting setup brought the experimenter remarkably close to the desired end goal. It demonstrated that powerful, multi-step coding workflows are not exclusive to proprietary systems; they can be built and replicated with open-source components.

This initiative highlights the exciting trend of local LLM deployment. It proves that users can take control of their AI development pipelines, fostering both privacy and deep technical understanding. Though there are still limitations, the journey of building these custom AI tools continues to push the boundaries of what is possible in software engineering.