The part between the model and the sandbox.
Why I built a routing and control layer for my local AI-agent workflow, how NVIDIA OpenShell fitted into the setup, and the limits I still have to account for.
Read the engineering note ↗Some projects become products. Others start because a tool does almost everything I need, except the part that makes it work in my own setup.
Why I built a routing and control layer for my local AI-agent workflow, how NVIDIA OpenShell fitted into the setup, and the limits I still have to account for.
Read the engineering note ↗Separating the engine, runtime adapters, storage and interfaces before writing another oversized application. A fresh Rust direction, not a rebrand of the earlier work.
Building CLI and browser-based control surfaces around runtime operations made me think harder about command parity, failure visibility and version-specific behavior.
A passing suite is useful, but the harder work is knowing which environment, dependency versions and live workflows the result actually covers.
Smaller experiments and systems work do not always need a release button. These are areas I've been exploring separately from my published applications.
A desktop-assistant experiment in Python, with a focus on local execution and tool-oriented workflows.
File and storage recovery exploration: scanning, filtering and thinking through what can reasonably be recovered.
Workflow tooling and integration experiments. Research and repository activity should not be mistaken for a supported public release.
Independent management and automation tooling around third-party sandbox infrastructure, with version-specific compatibility considerations.