mit

Attention over time

7-day rolling share of analyzed posts/comments mentioning this entity · 90 daily points ending 2026-10-15

Final complete day: 2026-10-15. Attention was 0.2% — this entity appeared in 6 of 2,852 analyzed posts/comments in the seven days ending that day (6 mentions).

Each point is a seven-day rolling window of complete UTC days; a point is comparable only when at least 100 analyzed posts or comments fall inside it. A day with no analyzed records counts as zero; missing coverage is marked and never shown as zero. The series ends at the latest complete analyzed day — it never extends into the current incomplete day. How attention is measured.

Recent mentions

1. 2026-10-12 17:34 UTCComment

   Yeah, that’s definitely one of the lessons I’m taking from this. The new version was already being developed separately as my own product, so fortunately the server only contained the older conference-specific version. And no, I didn’t release it under MIT or any other open-source license. Still, I agree that keeping my code and product development completely separate from an organization’s infrastructure would have avoided a lot of this.

2. 2026-10-10 18:08 UTCComment

   For anyone asking: it's all free (MIT). * npm: `@matita/icons` * Figma file: Matita icons on Figma Community * Iconify: `matita` Browse and copy:

3. 2026-10-10 16:44 UTCPost

   [Showoff Saturday] I made 173 hand-drawn-style SVG icons that are generated deterministically from their names - I wanted icons that look drawn with a pencil but behave like a normal icon set: 24 grid, `currentColor`, tree-shakable, and identical on the server and in the browser. So none of them are drawn by hand. Each icon is a few drafting primitives (line, polyline, rect, circle, arc…) on a 24 grid, and a small engine draws them freehand: lines overshoot their ends, rect corners cross, circles sweep past 360° and drift outward so they overlap where the pen closes, and each stroke gets its own pressure. The randomness is seeded by the icon name, so every build draws the same paths. That means no hydration mismatches and no extra runtime cost; the package ships the finished paths. The hard part was small sizes: all the wobble is capped (a line bows less than a pixel at 24 px), so it reads like a regular line set at 16–24 px and the pencil look shows from \~32 px up. * Browse and copy: * How the engine works: * Code (MIT): React, plain SVG strings, a script tag, or raw .svg files. Feedback welcome, especially on missing icons and how it looks in a real UI.

4. 2026-10-10 16:36 UTCPost

   theAuth: open-source auth where AI agents are first-class, not just API keys - **theAuth: open-source auth for AI agents and humans (TypeScript and Go), built solo** **theAuth** is auth for apps where AI agents act on behalf of users. Each agent gets its own token and permissions, a user can delegate access with limits, and every action is logged. It also covers the usual: **passkeys**, **SSO**, **orgs**, and an **OAuth 2.1** server so **MCP** servers can verify who is calling. By the numbers: * TypeScript core: 2,400+ tests, about 77% coverage, 3 runtime deps, 17 OAuth providers, 10 framework adapters * **Go SDK session lookup**: 58-67 us down to 30 us (about 2x, 33k lookups/s per process) * **Go SDK password sign-in**: about 75/s per instance, with hashing memory capped at 608 MB instead of 4 GB under a login burst * Those two are measured on an M4 Max with Postgres 16, 100k users, all on one laptop, and the load test script is in the repo * MIT, self hostable `npm install @glinr/theauth` runs on in memory SQLite with nothing else to set up. Go: I built this solo, with several AI tools and a lot of automation to keep a steady pace. Does the agent permission model make sense to you and maybe give it a try for future projects.. I already migrated all my 6+ websites to theAuth and been a great time saver !!

5. 2026-10-10 13:17 UTCPost

   [Showoff Saturday] I moved my screen recorder's rendering out of Electron into Rust: export went from 8 to 126 fps - OpenScreen is a free, open-source (MIT) screen recorder for product demos. I maintain it. The UI is Electron, React and TypeScript, and 2.0 just shipped. The part that might interest this sub is where the browser stopped being enough: - Rendering used to run in Electron, on PixiJS. Export topped out around 8 fps on our reference laptop (Ryzen 5 7520U, integrated GPU). - It now goes through a Rust compositor on Metal, Direct3D 11 or Vulkan, loaded as a native addon. Same effects, about 126 fps. - Capture moved to native helpers too: Swift on ScreenCaptureKit, C++ on Windows Graphics Capture, PipeWire on Linux. - The React UI stayed. It drives the compositor and owns the timeline, the transcript editing and the settings. On top of that: a CLI that renders a project from a script, and a local MCP server so Claude Code can edit the demo of the feature it just wrote.