In the past few weeks, experimenting with a custom ai agent running entirely in the browser with sandboxed tools exec.

The core is built in Rust, a native CLI is built on top for local experimentation but the most interesting part is the web version: the core is built to WASM and get augmented with many tools in the JS land: - OPFS access (read, list, edit files) - Sandbox Python exec (Pyodide in WASM) - Sandbox DuckDB exec (DuckDB-WASM) - Draw charts - Show images - etc OpenAI Completions compatible API providers are supported.

But if you want a full local and sandboxed execution of the whole agent, the web version bundles also wllama to serve local GGUF models (with WebGPU optional support).

Github repo: https://github.com/rclement/cooper

Very cool work. I've been seeking solutions for running an agent in the browser for building artifacts on my project, which is a web artifact tool library called Exhibit. I'm starring this for later.

https://github.com/momja/Exhibit

Oh nice! I’ll take a look at it, I was thinking of implementing such mecanism in Cooper (I don’t know where it’s going, exploring the possibilities and practical usefulness)

> People do weird things when given sudden access to money or power.

10 years ago my last boss told me one last advice before going onto entrepreneur ventures: « be careful, people do become crazy and stupid with money » (and I guess he knew what he was talking about…)

Same been running Debian 13 Gnome on MBP 2015 13’. Works great, in a triple boot setting using OCLP

For instance visio calls are unusable on Sequoia (OCLP) but rather smooth on Debian

I installed opensuse tumbleweed on mine and the bastid won't hibernate. Or: it hibernates but refuses to resume. It loads the saved state from disk and then 30 seconds of black screen later, reboots and loses state. Two solid days with Claude, exhausting every avenue, and it's basically unfixable. Sleep=OK, Hibernate=Fucked.

Is this using the youtube per-channel RSS feeds?

For people already using RSS feed readers, this is another way to achieve the same result (i do that with miniflux)

Yes, it is using RSS feeds per channel and also there are multiple fallbacks built. NoSuggest is a curated experience just like YouTube but without algorithm recommendations.

For the past few months, many times i’ve tried this workflow: 1. Ask a coding agent to think and implement a feature that is non trivial 2. This leads to really understand pros and cons for many possible solutions and see it happen end to end 3. Revert all changes and implement it myself when i’m settled on a solution i’m satisfied with 4. At this point the agent is just an iterative reviewer

I’ve felt that any non trivial amount of code not written myself tends to be hard to own. And like the author said, need to keep skills sharp also

Wow thanks TIL about the « add-bounds » config! Especially useful for project where pinning to exact dependencies is crucial and easily missed by less experienced devs (end products, not libs)

Because a coding agent (when instructed well) will try to make a piece of code work in a loop. Static typing and compilation help in the process (no more undefined variables discovered at runtime for instance). But that’s not bullet proof at all as most of us know

Might be the best visual explainer of Shazam original audio fingerprinting algorithm from the 2003 paper (I guess they´ve switched to ML models at some point?)

Agreed. But would mean having to educate people on security, privacy and computing in general… Pretty sure most government like having most people uneducated on such things

I feel like you could do it in an app or a card with an NFC chip.

People don't have to know security or cryptography to do their banking online.

Either way it would be infinitely better than the current social security number situation we have.

Agreed. Not saying there are no ways to ease the process but someone has to put some effort