Local execution
Build around Ollama so the developer controls where models run and where source code stays. The trade-off is that model quality and response speed depend on the local setup.
Local AI / Developer tooling
OwA
A coding agent built around local models, repository context, and a developer’s own machine.
Independent project · Zaka Noor
owa — workspace
$ owa › explain this deployment flow Reading the relevant code before answering…
01 / The problem
What needed to be clearer
A useful coding assistant needs to understand the repository it is working in. For a local tool, that understanding also has to fit the constraints of smaller models and local hardware.
02 / The approach
How I approached it
OwA brings repository search and tool use into a command-line workflow built for Ollama. The portfolio’s implementation focus is grounding answers in relevant code while keeping the workspace on the developer’s machine.
Repository context
Use hybrid search to find relevant code before answering. This makes retrieval a core part of the workflow, alongside the model itself.
A focused interface
Use a CLI that fits the development workflow, with explicit tools for working with repository context.
03 / The result
What you can explore
An open-source local coding tool, with the implementation available for inspection. The repository is the place to check current capabilities, setup instructions, and limitations.
Have a related problem?