W42 This Week's Picks

Solves the information gap where buyers cannot get real tok/s before buying local inference hardware, turning measured results across hardware combos into interactive comparisons. 175 points come from the neutral-data vacuum amid a consumer GPU upgrade wave; exploding in two days validates demand. Path: hardware vendor partnerships, procurement consulting or enterprise POC data services. Risk: official llama.cpp benchmarks and OpenRouter-style leaderboards dilute it, and data credibility may be questioned.

#LLMs#Benchmarks#GPU

Solves the problem that shared windows in pair programming can only become grid tiles: shared windows exist as real windows you can click into, type and scroll. After Screenhero was shut down by Slack and CoScreen by Datadog, a vacuum appeared; the original team rewrote it in eight weeks and promises never to shut it down, and 102 Show HN points plus founder backing is strong momentum. Hosted instances and self-hosting are the monetization path, but the no-charge pledge is the biggest uncertainty, with Tuple and VS Code Live Share as rivals.

Hi HN! I’m Jason Thomas, one of the founders of CoScreen. As some of you may know, Datadog sunset the CoScreen service on July 31st. Four of us from the original CoScreen team have spent the last coup…

#Collaboration#Productivity#Open-Source Alternatives#Video#Self-Hosted

A free, open-source Mac dictation tool aimed directly at paid products like Wispr Flow, with 140 points in half a day proving voice input is a real need with many underserved users. Open source speeds adoption, then team editions or local models could charge; the risk is Apple's system-level dictation improvements and the incumbent's brand.

#Open-Source Alternatives#Speech#macOS#wispr-flow

Statically recompiles PS2 games to WASM to run natively in the browser, technically harder than a normal emulator; 260 Show HN points is the strongest momentum in this batch, from nostalgic players and the reverse-engineering community. It could become a PS2-to-WASM porting pipeline service for porting studios, but Sony copyright and ROM compliance are hard risks, and a demo alone is unlikely to become a standard dependency.

#Game Porting#WebAssembly#Reverse Engineering#Games

Press a hotkey to talk; simple tasks finish locally via rules in 0.08 seconds, complex tasks go through Claude -p without an API key, and successful actions are saved as replayable scripts to cut costs; 55 Show HN points and a young author's story boosted spread. Commercialization could be subscription or enterprise, but it faces Raycast AI, Siri, and the macOS 26 barrier, and a personal project has a thin moat.

Hi HN! I'm Saarth. I'm in 9th grade, and I built Atticus. Atticus is a fast and cheap Mac assistant that can do tasks for you. You hold down Control-Option-Space, speak or type your task, be it comple…

#macOS#Speech#AI Agents#Productivity

SolidCraft is an open-source parametric CAD tool in Rust with MCP support, and 133 Show HN points show high community expectations for AI-drivable CAD. CAD has long been dominated by SolidWorks and Fusion 360, so an open-source alternative is itself a chance to become a standard dependency, and MCP lets it plug into agent workflows. The risk is that parametric CAD kernels are extremely hard to build deeply, and mature competitors are far ahead in features.

#3D#AI Agents#MCP#Open-Source Alternatives

Agor focuses on multiplayer AI, putting team members and agents in the same collaboration space, and 1,446 GitHub stars show multi-agent human-AI collaboration is a real current need. Commercialization could be team collaboration SaaS or an enterprise edition. The risk is that orchestration frameworks like LangGraph and CrewAI are all moving toward collaboration, and only 9 HN points suggests broad awareness hasn't opened up yet.

#AI Agents#Collaboration#AI

Has LLMs play StarCraft BW by writing code, essentially a public benchmark for agent long-horizon planning and API operation; 233 points come from the natural spread of model-arena topics. Path: become the standard leaderboard cited by eval media and vendors, taking sponsorships, API distribution and consulting. Risk: game benchmarks age fast, and aggregators like OpenRouter building their own leaderboards dilute citation value.

I've refreshed StarSkirmish Bench, a benchmark where LLMs play StarCraft BW by writing code, with results from the flagship models of OpenAI, Anthropic, Google, and Muse

#Benchmarks#AI Agents#Games#LLMs

Fills the gap where local event data (pop-ups, shows, one-off events) has no unified supplier, exposed directly to agents as MCP, using influencer accounts for fan-out aggregation faster than real-time crawling. 32 points show the pain is real but reach is limited; the author is validating paid demand. Path: become the event data API for the agent era, benchmarked against Eventbrite data services. Risk: narrow coverage, SF-only, easily copied by big tech or ticketing platforms.

I built an api/mcp for local events: pop-ups, one-off events, concerts, etc. currently covering the SF bay area. I've been building a bunch of consumer apps lately around personalization and recommend…

#MCP#APIs#AI Agents#Data Analysis

Turns the agent research process into a shared lab, solving the silo problem where agent exploration results in a team cannot be reused or audited. 54 points show multi-agent collaboration is moving from toy to real need. Path: a subscription collaboration platform for engineering teams, like a shared notebook for agents. Risk: observability platforms like LangSmith and Weights & Biases casually absorbing this.

#AI Agents#Collaboration#Knowledge Base