New week, new Rust! What are you folks up to?
Hii, I am working on a simple drawing software called Pixeldraw.
I'm still working on my DCE-inspired middleware, where I implemented something I call "live sessions", which allows the client to group calls. It occurred to me that I always end up implementing this thing when I build more complicated applications using middleswares, and it's never as robust as I want it to be. Turns out that with native support in the middleware, it's much more robust and much simpler to use.
I'm hoping to finish off the "graceful shutdown" in the "low-level" network crate that the middleware uses.
I watched Logan Smith's video on the "perfect function" and it made me realize that I really should split up my functions even more. So I did, and it made reasoning about the shutdown much simpler.
I also wrote a bunch of tests for my block-level deduplicator. This week I'm going to write a test-application for it, to see if the API needs any tweaks.
I watched a video about the Verse programming language and became a little obsessed with effects systems, and am going to spend some time reading up on what's going on with effects in Rust.
I’m working on psp: https://github.com/MatteoGuadrini/psp
This is my principal project since two years.
I'm still tinkering with Nuillu and finally published a web demo at nuillu.com (powered by egui). Nuillu is both a cognitive agent and a generic runtime for building specific kind of concurrent multi-agent systems.
Next, I'm planning to experiment with an octopus-like architecture: eight (or more) subsystems, each with its own attention and cognition. Not just to role play an octopus really well, I want to explore whether this kind of distributed cognition can help an agent work on large, complex tasks.
I'm building a headless browser for AI agents — one Rust binary with embedded V8, a Rust HTTP stack, and a CPU-only CSS/layout/paint renderer (Blitz/Stylo/Taffy lineage as the reference implementation) so it can screenshot without Chromium. Agents drive it over HTTP/MCP/CDP: read pages as markdown, render JS/SPAs, screenshot, search, and interact through persistent sessions.
The renderer has been the most interesting part: layout through taffy, text through a bundled CJK font subset (GB2312 + symbol blocks) where I learned to check actual raster ink rather than trusting the cmap. Most "agent browsers" are thin wrappers around Chromium; I wanted to see how far a self-contained binary could go. It's open source at yinnho/aginxbrowser if anyone's curious.
I've started work on a prototype language analyzer for canonical Pāli using Tantivy. I have a custom tokenizer working:
I'm working on a Stream splitter (a sort of reader.read_until(fn)). It's not as simple as I initially thought.
Currently it consumes 3 times more CPU time than a minimalist algorithm. I need to simplify my overall algorithm to reduce the number of conditions that handle just 5% of cases.