One thing I don’t cover in the blog post is ensuring your MCP tool calls are well documented. (If it’s an existing MCP you can do this with a README or instructions file.) I saw a jump in efficiency when I manually edited the docstrings with examples of when each tool would be used, how to call it, and better argument descriptions.

That's cool! Have you thought about some docstring augmentation loop -- either by having the agent log the tool uses or by auto querying (e.g. perplexity API) for example use cases in the wild?

Would be forever grateful for a repo to learn from, even if its rough.

Author here, I think you’re spot on. If I set out specifically to make a mashup and spent the same number of hours working toward that goal I think I could have gotten to the same place. This was a journey that started with wrapping my head around MCP and LLM local software interoperability, and my Ableton knowledge leveled up a lot along the way.

I think this tooling could be useful in the hands of more capable musicians / audio engineers / etc. as there are often repetitive tasks in DAWs and it could potentially unlock new workflows that would have been too tedious without knowing how to program.

I think it just got folded under the umbrella concept of model alignment. And it moved from theoretical discussions to practical daily struggles with LLMs deleting failing unit tests

The Sumerians were on to something: https://en.m.wikipedia.org/wiki/Sexagesimal

In engineering, they used to use the foot ' inch '' line ''' and point '''' where each unit was 1/12th the size of the previous one. The european typographic point used to be 1/144th of an inch. Watch components are measured in points.

That’s exactly what I have. The C++ codebases I work on build against a specific pinned version of LLVM with many warnings (as errors) enabled, and building with a different version entails a nonzero amount of effort. Ubuntu will happily install several versions of LLVM side by side or compilation can be done in a Docker container with the correct compiler. Similarly, the TypeScript codebases I work with test against specific versions of node.js in CI and the engine field in package.json is specified. The different versions are managed via nvm. Python is the same via uv and pyproject.yaml.

I don't doubt it, but I don't think that situation is accepted as the default in C/C++ development. For the most part, I expect OSS to compile with my own clang.

I didn’t see any mention of running a model locally on device like the Hibiki abstract states. Is this available?

three.js might not be the right fit for computational rendering/raytracing, optimization techniques like frustum culling are more or less baked into the engine.

Was https://hyperaide.com/pricing added after your comment?

Nope, was always there. Not sure what they mean.

For low dimensional (such as 2D) projections of high dimensional data, especially useful for visualization, take a look at UMAP - https://umap-learn.readthedocs.io/en/latest/

vscode remote containers are still the standard, but I find them very usable nowadays. My setup is a MBP M2 that I use to remote into a Windows WSL setup at home, a Linux desktop at work, and various servers. Nesting remote SSH + remote Docker works seamlessly, that was previously a major headache.