OpenScreen 2 finished #5 Product of the Day on @ProductHunt ๐
Your turn to get seen: post the demo you made with OpenScreen in #made-with-openscreen on our Discord. Self-promo welcome.
Every Monday we feature a Demo of the Week here ๐
https://t.co/2CTRpmG9RM
@0xwhrrari Same pattern works inside an editor, on real footage.
In @getopenscreen, Opus 5.5 writes the effect as a GPU shader, it lands on the timeline with its own sliders, and it checks its own frames before stopping. Same pixels in preview and export.
Write-up: https://t.co/dbEO4y8n8Z
OpenScreen 2.1 is out, and it's a big one.
Two things change what you can ship:
โ demos made by your agent, attached to your PRs
โ virtually unlimited visual effects: if the menu lacks one, your agent writes it
Free, open source. Windows, macOS, Linux.
OpenScreen now works as a plugin in the ChatGPT desktop app.
Add the marketplace getopenscreen/openscreen, install the plugin, then ask it for a demo video in one sentence. It plans a timed storyboard first, then records, edits and exports the MP4.
@OpenAIDevs
@malmgrenkevin@alexandr_wang Built one: MuDesktop. Projects, a review panel, terminal and browser, worktrees, MCP, opt-in computer use. Free, open source, in the Microsoft Store and notarized on macOS.
@mjdouglas MuDesktop runs whatever Muse Code you have installed, so 1.4.8 works in it as is. A desktop app for Muse Code: projects, review panel, worktrees, MCP, opt-in computer use. Now in the Microsoft Store and notarized on macOS. Free and open source.
OpenScreen is now a plugin in the ChatGPT desktop app.
Add the marketplace getopenscreen/openscreen, install the plugin, then ask it for a demo video in one sentence. It plans a timed storyboard first, then records, edits and exports the MP4.
@OpenAIDevs
I built MuDesktop, the desktop app for Muse Code, Meta's coding agent.
Projects, live tool calls, a review panel, terminal and browser, git worktrees, MCP, and opt-in computer use. Free and open source, on Windows and macOS.
An independent project, not affiliated with Meta.
Trending in 15 s: Diagram Design, an agent skill for editorial diagrams.
One diagram in three styles, then the exploded AI agent stack replays layer by layer. Self-contained HTML + SVG, no Mermaid slop.
@cathrynlavery
Harnessยฒ, from Google Research: keep the model fixed, improve the harness around it, recursively.
Propose edits, probe them on the same task, compose a better harness, then learn how to improve the next one. On one JobBench task: 43.1 โ 77.8.
@GoogleResearch
DeepSeek-V4.1-Flash-1017 in DeepSeek Harness: one request, sped up.
It plans, then turns a bare page into a trail finder: search, difficulty filters, sort by distance, dark mode. Plain HTML, CSS and JS. The status bar reads ~300 tok/s.
@DeepSeekHarness@deepseek_ai
Trending in 15 s: Anima, interactive biology figures made by Claude.
Hover a receptor and it explains itself, slide the dopamine level, block D2 and watch the signal flip. Then a firing neuron whose numbers come with their sources.
@Zea1shama
Trending in 15 s: MonoCode, a GUI for your coding agents.
Its Monos are agents that work across your projects: one request, the Mono starts a session in the project, then reports back with the changes.
@nikolay_dp
Trending in 15 s: llmspeed, feel local LLM speed before you buy.
Same model, same prompt: RTX 4090 1.8s vs MacBook M4 Pro 5.0s. Then a Mac Studio M5 Ultra takes the next race in 1.5s.
@nicorobinreal