This past weekend @fhackdroid and me got an opportunity to talk at IICT 2026 @compiler_tech hosted at @iiscbangalore on waspy, a python to wasm compiler, our progress so far, things that work now, and things being worked upon... 🎉

This uses current state of waspy to compile python to wasm, so not all cpython modules are currently supported... You can see all the modules supported currently here: anistark.github.io/waspy/mod… ...and we keep this updated as we work more on it... This is based on py 3.12 so perhaps we might be missing some modules, but so far, this is the plan till v1 🔥

We've a new Python to WASM playground on waspy website now, where you can try diff python programs and see it's WAT, as well as functions to interact with. First presented yesterday at CompilerTech conference at IISC, Bengaluru with @fhackdroid Come check this...

When LLM doesn't understand what "realistic" means...

Haiku still looking for invite to the 5.5 party 🥲

"Personalisation" in every products is gonna take a giant leap forward.

Had a great time talking about building 3D worlds… So great to find folks working on incredible stuff in the area… at @IndiaFOSS 2026 ⚡️

Heya folks Got new updates from feluda v1.17.0 ✨ We worked on letting Feluda look inside container images. Save an image to a file with docker save and Feluda tells you the licenses of everything in it. The OS packages and the libraries baked into Go programs. It also works offline now, since you can give it a file of license answers when there's no internet. The interesting part is that you don't need Docker, Go or any database tools installed to scan an image. Feluda reads the image file, the Go programs and the package databases on its own.

Really cool work on CUDA oxide by Nvidia Labs ✨ Now, I wanna learn more about GPU programming ⚡️

Reading is the new bottleneck…

I think everyone should build their own personal assistants at this point. :3

wasmrun v0.22.0 is now out 🎉 `wasmrun agent` gives an AI coding agent an isolated sandbox to run code in. No Docker, no daemon, no browser. Just a REST API over a WebAssembly interpreter. Snippets were already supported, but v0.22.0 adds support for projects as well. An agent could write TypeScript but couldn't run its tests. And when something threw, the stack trace pointed at transpiled JavaScript the agent had never seen. Both are fixed. A thrown error now names the line you actually wrote: Error: kaboom at boom (lib/thrower.ts:4) at <anonymous> (main.ts:3) That matters more than it looks. An agent that can't map a failure back to its own source can't fix it, it can only guess. Try it out: cargo install wasmrun Release: github.com/anistark/wasmrun/… Docs: wasmrun.readthedocs.io/

sot has an excellent cleaner 😛 but feels incomplete now without all the agent caches. Maybe time to add to it... `sot clean`

The `--gui` mode got a complete upgrade. New table layout, a cleaner theme, and smoother interactions for browsing every dependency and its license in one view. Scanning a big project and reading the results now feels like a real tool, not a wall of text. Run: **feluda --gui**

feluda v1.15.0 is now out 🎉 Most license scanners only read your manifest. So anything copied, vendored, or pasted straight into your repo stays invisible, licenses and all. feluda v1.15.0 now reads licenses from your actual source code, and a lot more. 🧵

The doomsday clock’s running now… ⚡️

wasmrun 0.21.0 is out. 🎉 Agent mode runs TypeScript and npm packages now. An LLM writes normal TS, imports lodash, POSTs it to `/exec`, and it runs inside an isolated WASM sandbox. No node on the host, no container per request. The transpile happens inside the sandbox too, by an swc build compiled to WASI. That keeps the single binary and the isolation model intact, at around 40ms per transpile.

Anybody still uses PHP in production?

For a weekend project, this became quite big 😅 I'm creating an open world of One Piece on the web ✨ Started with Foosha village... 🏴‍☠️ You can walk around foosha village or sail the east blue right now. ⛵️ Hopefully, will have more interactive elements going forward...

Give the model tools and Orion runs the whole loop: it parses the call, executes your tool, feeds the result back, and repeats until the model has a final answer. Every token and every tool call streams out as an event your UI subscribes to. Install : crates.io/crates/orion-core

The hard part of any agent is fitting a growing conversation into a fixed context window. Orion prunes to your token budget for you, either a sliding window or a running summary, and lets you pin the messages that must never get dropped. orionpod.com/#orion-core

Orion is backend-agnostic. You implement one trait, LlmBackend, with three methods: generate, tokenize_count, is_ready. llama.cpp, MLX, ONNX, a cloud endpoint, all the same to it. Your engine runs on a blocking thread. Orion owns the async around it. anistark.github.io/orion-cor…

Running a local LLM is basically a one-liner now. Building a real agent on top of one is not. You still owe yourself context pruning, token budgets, a prompt template per model, a tool loop, and streaming. I built Orion Agent Harness as an open source project to facilitate just that, written in rust. 🦀👇

Also shipped: a revamped landing page with an interactive architecture flowchart, so you can see the compiler pipeline end to end. crates.io/crates/waspy Try it, break it, open issues.

After about 6 months break, waspy v0.10.0 is out 🎉 This one's mostly fixes... The WASM it emits now actually does what your Python says: floats, collections, strings, classes, control flow. 👇

Chrome isn't secure to use Gmail ? 🤔 @googlechrome @gmail @Google

Some hashiras take vacation ✨

Dropped a short post on how to Surface License Violations in GitHub Advanced Security with feluda 🦀 which includes the recent addition to Static Analysis Results Interchange Format (SARIF) support. 👇 Currently supports SARIF 2.1.0 specs ✨

You and me both spidey…

Spider-Noir is on 🔥 🕷️

Listen Robin… #OnePiece

Feluda v1.13.0 "Hatyapuri" is out 🕵️ We got SARIF, monorepo, init, claude plugin and more 👇

Node.js runtime build infrastructure in now available on wasmhub v0.2.0 🚀 Uses QuickJS as the initial JavaScript engine since it compiles cleanly to WASM/WASI as of today. ⬢ ✨ 👇

Say hello to Pixelspace 👋 An original 5×7 pixel display typeface. Single weight, 474 glyphs. Covers the full Google Fonts Latin-core set (A–Z, punctuation, Latin-1, Latin Extended A/B, smart quotes, currency, combining diacritics) plus Unicode Box Drawing and Block Elements, so TUI borders stitch and btop bars LEGO-tower. Open Source and hackable: the source is a plain bitmap text file (. off, # on). OFL 1.1. Free for personal and commercial use.

Truly living in a time where only imagination is your limit...

ono... its a tui components library XD 🎉 Goal is to build @ratatui_rs components to be used as is in your TUI projects. Supports themes and some basic components right now. Will build more ✨

Wrote about writing your own claude plugins. Can be skills, hooks, commands, MCPs or LSPs. Start from writing to distributing it to everyone to use... ✨

Is it just me or everything’s moving way too fast… Are you even breathing?

First thing I did with @claudeai Design XD

Map Concept...

Dashboard Concept...

Playing around with TUI styling using @ratatui_rs ✨ Which theme do you like best? Comment A, B or C below 👋

`sutras validate` now accepts a path to a skill directory in addition to a skill name, so projects with non-default layouts (e.g. plugin repos with skills under `skills/`) can validate without symlinks or custom discovery. 🎉 🚀 Upgrade to sutras v0.4.5 Tested py cli with opus 4.7 ✨ and also made available for pi-extension on npm 🥂

Amaze, amaze, amaze! 👎

How about CLI System Obervation Tool 😁 github.com/anistark/sot

Your plugin code stays in your repo Your license, Your rules The marketplace only stores references One PR to list yours One command to install anything 🎉 no accounts. no review queue. no lock-in Try it out on your @claudeai code now:

Claude code has a plugin ecosystem that's growing really fast. ⚡️ But actually finding plugins? that's a different story... There's no central place. no search. no catalog. just random repos, discord links, and hoping someone tweets about it. There's some awesome plugins repos as well, but do you go through them and browse each repo? How about browsing from your claude code itself, with a community marketplace... ✨

Playing around with some UI updates on my blog. Was thinking about how every website has dark/light themes only. But most of the times, I'd want a slightly diff color scheme. Added a bunch of themes here. Let me know what you think

How many reviewers does your PR actually need? Made a GitHub Action that figures it out automatically. Analyzes change size, weighs file types differently, and labels your PR with the ideal reviewer count. Zero config. Just plug and play.

Unreal, unprecedented speed of releases and updates. Product lifecycle has changed forever. Anybody building on older theories is just going to fall behind. This isn't software moving faster. Intelligence itself has become a primitive and every business built on the old assumptions needs to rethink from the ground up. Wrote about what that actually looks like 👇

"I'm not alone, when I'm alone" -- Tommy Shelby

Most agents today are being built the same way we once wrote our first prompts. Experimentally, intuitively, and often in isolation. It works, until it doesn’t. But building reliable systems has never been a mystery. We’ve spent decades refining how we design applications: layering abstractions, defining boundaries, planning for scale, and embracing extensibility by design. Agents don’t require a new philosophy. They require a translation. The real shift isn’t from “code to prompts,” but from applications to agents, carrying forward everything we already understand about architecture, and rethinking it through a new interface.