25-05

Tab cleanup

• Wanix :: Virtual environment kit for the local-first web, inspired by Plan 9 | Lobsters

• Show HN: Magnitude – open-source, AI-native test framework for web apps | Hacker News

• Show HN: I used OpenAI’s new image API for a personalized coloring book service | Hacker News

• Open-source interactive C tutorial in the browser | Hacker News

• HTTP Feeds | Asynchronous event streaming and data replication with plain HTTP APIs.

  • Asynchronous event streaming and data replication with plain HTTP APIs.

• cloudevents/spec: CloudEvents Specification

• The Coming Knowledge-Work Supply-Chain Crisis

• pabateman/kubectl-nsenter: kubectl plugin for pod’s linux namespaces command execution via direct node ssh connection

  • exec to any pod’s container linux namespace

  • uses a direct connection to node via ssh

• Against Horizontal Scroll

• FAQ | Pyrefly

  • A faster Python type checker written in Rust

• I use Zip Bombs to Protect my Server | Lobsters

• Systems Performance 2nd Edition Book

  • book by Brendan Gregg

• $ Shell – API | Bun Docs

  • Bun = JS/TS runtime with extras

• Inkeep - AI for your users and support team

  • Custom Chatbot

  • “Ask AI”

  • ycombinator

• Making PyPI’s test suite 81% faster - The Trail of Bits Blog

  • pytest-xdist

  • sys.monitoring (note: no branch coverage)

  • testpaths

  • pytest-sugar for nicer output

• Bypassing a web application firewall with autocomplete | Lobsters

• Obtainium

  • install Android apps directly from Github/Gitlab/… releases

  • via GitDroid: A third party Android app manager for apps uploaded to GitHub releases | Hacker News

• ePub-utils: A Python library and CLI tool for inspecting ePub from the terminal | Hacker News

  • alternative: EbookLib

  • mupdf / PyMuPDF

• Jacksaur/Gorgeous-GRUB: Collection of decent Community-made GRUB themes. Contributions welcome!

  • grub themes

• zerowidth positive lookahead | jj tips and tricks (jujutsu)

• A Better Man Page Viewer - Notes from VisualMode

  • neovim as manpage viewer

  • “nvim manpager fuse: mount failed: Permission denied” neovim/neovim#15590

  • export MANPAGER='nvim --appimage-extract-and-run +Man!'

  • via

• Nari Labs: Dia Examples

  • nari-labs/dia

  • not available on OpenRouter

  • Dia 1.6B ZeroGPU space

• Show HN: Real-time AI Voice Chat at ~500ms Latency | Hacker News

• OpenRouter Model Price Comparison

  • Show HN: OpenRouter Model Price Comparison | Hacker News

  • OpenRouter Structured Outputs Playground

• akinomyoga/ble.sh: Bash Line Editor―a line editor written in pure Bash with syntax highlighting, auto suggestions, vim modes, etc. for Bash interactive sessions.

  • tried it. too slow for my taste

• A Critical Look at MCP - Raz Blog

• Ask HN: What are good high-information density UIs (screenshots, apps, sites)? | Hacker News (usability, ux, ui)

• git tools

  • Mergiraf better git merge

  • Difftastic — Rust application // Lib.rs diff based on AST (vs. delta)

• 🐌 Slow Software for a Burning World 🔥

  • fediverse

  • slow software philosophy

  • counter point to “move fast and break things”

• Gmail to SQLite | Hacker News puddl

• Brandon’s Semiconductor Simulator | Hacker News

• Software License Management with Polar.sh | Lobsters

  • Polar — Payment infrastructure for the 21st century | Polar

• Wtfis: Passive hostname, domain and IP lookup tool for non-robots | Hacker News

• OpenTelemetry protocol with Apache Arrow | Hacker News

  • row-vs-columnar.png (PNG Image, 1624 × 1414 pixels) — Scaled (68%) nice visualization of row vs column layout in databases

• If AI is so good at coding … where are the open source contributions? – Pivot to AI Mirrors what I feel about AI tools. For now, they are good at writing boilerplate, but weak at engineering.

• Experts have it easy makes some good points about novices and experts with a maze analogy. It’s about learning and teaching.

• Show HN: VectorVFS, your filesystem as a vector database | Hacker News

• The Bitter Lesson (2019) | Lobsters says that brute force search and learning proved to be superior to encoding human knowledge for machine learning.

• The State of SSL Stacks by HAProxy shows that OpenSSL v3 is really slow and benchmarks other implementations (BoringSSL, LibreSSL, WolfSSL, AWS-LC)

  • In blunt terms: running OpenSSL 3.0.2 as shipped with Ubuntu 22.04 results in 1/100 of WolfSSL’s performance on identical hardware!

  • It’s also visible that a 32-core system running optimally at 63000 connections per second on OpenSSL 1.1.1 would collapse to only 1500 connections per second on OpenSSL 3.0.2, or 1/42 of its performance, for example, after upgrading from Ubuntu 20.04 to 22.04.

• Launch HN: Exa (YC S21) – The web as a database | Hacker News

  • web search with tabular output

• Write the most clever code you possibly can | Lobsters asserts that writing “clever” code results in harder debugging, which forces you to learn. It makes a counter-point to Kernighan’s quote. As always: It depends. ;)

• Commodore 64, 1982 - commodore.commodore64.1982.102646264.pdf C64 ad from 1982 ♥️

• Polaro is an electric keyboard with big display and some “ai” stuff. Could be useful for teaching piano remotely.

• Plain Vanilla Web | Hacker News

• Why Bell Labs Worked | Hacker News

• What is HDR, anyway?

  • with HDR you loose detail, because you overlay multiple images.

• Launch HN: Tinfoil (YC X25): Verifiable Privacy for Cloud AI | Hacker News ai, trusted computing, enclave

• If AI is so good at coding … where are the open source contributions? | Lobsters makes a good point that shows that current LLMs are not yet able to do software engineering

• A Ledger In PostgreSQL Is Fast | Lobsters pgledger can run about 10k (10e5) transactions per second on a laptop

• Bridging the Gap Between Keyword and Semantic Search with SPLADE - Arcturus Labs shows how to use SPLADE with Elasticsearch

• Show HN: Airweave – Let agents search any app | Hacker News wraps tools (asana, google drive, sql DBs, custom) as searchable REST API or MCP server

• Push Ifs Up and Fors Down | Hacker News programming talk. I like batching.

• vrtmrz/obsidian-livesync I don’t know if I want to use obsidian, as I am happy with my sphinx based setup, but it’s good to know that there is free sync.

• Xata: Postgres with data branching and PII anonymization | Lobsters interesting concept

• Immediate-Mode Graphical User Interfaces - 2005 - YouTube for me is a lesson on how designing state correctly (hot / active UI element) can massively reduce the amount of work that needs to be done. Another example that comes to mind is Flask ApplicationContext and RequestContext.

• freezed | Dart package “yet another code generator for data classes, tagged unions, nested classes and cloning”

• pgvector for Python developers | POSETTE 2024 - YouTube nice intro to pgvector

  • slides pgvector for Python developers

  • pamela fox’s talks

  • via

    • pallets-eco/flask-sqlalchemy#1140

    • pamelafox

• Loading Pydantic models from JSON without running out of memory

  • via Loading Pydantic models from JSON without running out of memory | Hacker News

  • I would use DuckDB 😅

• Devstral | Hacker News maybe worth self hosting

• Building a fast website with the MASH stack in Rust | Lobsters

• matcha.css | Drop-in semantic styling library in pure CSS

  Ideal for fast prototyping, static HTML pages, Markdown-generated documents, and developers seeking to streamline their workflow without delving into CSS intricacies and want to make use of the full range of available HTML elements.

  • via The HARM Stack (HTMX, Axum/AlpineJS, Rust, Maud) Considered Unharmful | Thanh’s Islet 🏝️

  • via Building a fast website with the MASH stack in Rust | Lobsters

• sgl-project/sglang: SGLang is a fast serving framework for large language models and vision language models. inference engine

  • via Look Ma, No Bubbles! Designing a Low-Latency Megakernel for Llama-1B · Hazy Research

  • PyTorch Landscape via SGLang Joins PyTorch Ecosystem: Efficient LLM Serving Engine – PyTorch

  • skypilot-org/skypilot: SkyPilot: Run AI and batch jobs on any infra (Kubernetes or 16+ clouds). Get unified execution, cost savings, and high GPU availability via a simple interface. via Install SGLang — SGLang

  • Matching CUDA arch and CUDA gencode for various NVIDIA architectures - Arnon Shimoni search for sm75, because flashinfer only targets sm75+ as per Install SGLang Common Notes

    • my GTX 1080 is sm60 aka. Pascal (CUDA 8 and later)

    • i cannot run flashinfer :(

• AMD is propagating vllm for qwen3 without flash attention https://rocm.blogs.amd.com/artificial-intelligence/qwen3-day0-amd/README.html

  • I have no idea what flash attention is, but FP8 vs FP16 makes a huge difference according to the graphs here: https://triton-lang.org/main/getting-started/tutorials/06-fused-attention.html

• “An app is just a website that we have wrapped in the correct DRM to make it a felony to protect your privacy while you use it.” Cory Doctorov at the Pycon US 2025 keynote

• Pycon US 2025 Lightning Talks - Sunday, May 18th, 2025 AM - YouTube

  • cat emoji in sphinx and reveal.js 😅 https://youtu.be/lXngPPRaqGg?list=PL2Uw4_HvXqvb98mQjN0-rYQjdDxJ_hcrs&t=1018

  • t-strings https://youtu.be/lXngPPRaqGg?list=PL2Uw4_HvXqvb98mQjN0-rYQjdDxJ_hcrs&t=2291

• more pycon us 2025 talks

  • spawn vs fork and threads https://youtu.be/Uuhu-F05A7k?list=PL2Uw4_HvXqvb98mQjN0-rYQjdDxJ_hcrs&t=2306

  • csnakes: Python in C# .NET https://youtu.be/Uuhu-F05A7k?list=PL2Uw4_HvXqvb98mQjN0-rYQjdDxJ_hcrs&t=2849

  • advanced-alchemy https://youtu.be/47NJt_3ED_k?list=PL2Uw4_HvXqvb98mQjN0-rYQjdDxJ_hcrs&t=127

  • ckan DMS (data management system) for powering data hubs and data portals https://youtu.be/47NJt_3ED_k?list=PL2Uw4_HvXqvb98mQjN0-rYQjdDxJ_hcrs&t=799

• llama3-groq-tool-use with tool use for zed

• InternLM/lmdeploy: LMDeploy is a toolkit for compressing, deploying, and serving LLMs. vllm alternative?

  • via vLLM v0.6.0: 2.7x Throughput Improvement and 5x Latency Reduction | vLLM Blog

• Buying a GPU for LLMs in March 2025? Read This First! | Hardware Corner

• I Will Fucking Piledrive You If You Mention AI Again — Ludicity as some strong opinions language about ai

• Black Forest Labs FLUX.1 Kontext

• Rive animation toolkit

  • via Lottie is an open format for animated vector graphics | Hacker News

• svsc: Why? hynek, Python, service locator, dependency injection, inversion of control

• FireDucks Compiler Accelerated DataFrame Library for Python with fully-compatible pandas API

• Why you shouldn’t use vector databases for RAG Meilisearch makes the case for RAG with Hybrid Search sandwiched in LLM calls

Speech to Text (streaming whisper)

• https://www.reddit.com/r/LocalLLaMA/comments/1d1j31r/faster_whisper_server_an_openai_compatible_server/ mentions faster whisper server

  • nowadays called speaches: speaches-ai/speaches

  • integrates with Open WebUI (open-webui/open-webui#2248)

• Open WebUI

  • https://docs.openwebui.com/features/

    • https://n8n.io/ via ad for n8n –> n8n-io/n8n

      • hn? https://news.ycombinator.com/item?id=43879282

        • tldrsec/prompt-injection-defenses: Every practical and proposed defense against prompt injection. via https://news.ycombinator.com/item?id=43883317

External IDs

• postgres

• next to a PK

• ideally libs in python, js, C

• like ASINs

• Identity Crisis: Sequence v. UUID as Primary Key — brandur.org

• geckoboard/pgulid: Universally Unique Lexicographically Sortable Identifier (ULID) for PostgreSQL

  • a little long