Notes

@zachmeyer built Locker, an open-source Dropbox/Google Drive alternative that stays storage-provider agnostic. The interesting bit is the architecture: bring your own bucket, mount S3/R2/Vercel Blob/local storage behind a virtual filesystem, and layer search on top instead of treating sync as a monolith. Useful direction for teams that want file collaboration without platform lock-in.

@irushi points to Ares MBL, a prompt-layer guide for making models like GPT-5.4, Gemini, Ollama, and Codex behave more like Claude. The repo names eight recurring failure modes — from sycophancy and verification avoidance to completion theater and context amnesia — and packages the lessons into a drop-in system prompt.

Zero 1.0 — @aboodman and Rocicorp declare Zero stable after two years of work and 50+ releases. Zero is a sync engine for local-first apps that handles real-time collaboration and offline-first data without the complexity of building your own sync layer.

TurboQuant from @GoogleResearch — a compression algorithm that reduces LLM key-value cache memory by at least 6x and delivers up to 8x speedup with zero accuracy loss. The KV cache is often the memory bottleneck in long-context inference, so this could meaningfully extend context windows on existing hardware.

After WIMP — @threepointone explores what happens when users can describe procedures in natural language that compiles to code. The historic split between programmers (who reshape the machine) and everyone else (who use pre-built interfaces) starts to dissolve. Every app becomes directly programmable on demand, not through menus but through conversation.

Playbit — @rsms and team release a new runtime for personal-scale software. A minimal ABI-stable syscall interface that lets you write highly dynamic, collaborative graphical apps once and run them anywhere. Designed for programs by you, for you and the people in your life — filling the gap between web apps and native development.

@connors introduces Sutro — an email client designed for people and their agents. Bring your own Claude, Codex, or OpenClaw agents and have them work with you directly in your inbox. Email is not going anywhere; it is just getting agentic.

@jarredsumner's approach to Bun stability: a JavaScript engine fuzzer runs against Bun and Node APIs, and when it finds bugs, it automatically spawns Claude to fix them. A tight loop where automated testing surfaces issues and AI handles the remediation — no human bottleneck in the bug-fix cycle.

@aidenybai releases Expect — open-source CLI that lets agents test your code in a real browser. Point Claude Code or Codex at your app, get video recordings of every bug found, then fix and repeat until all tests pass. Works as a standalone CLI or agent skill.

cookie-es v3 from @pi0 and the unjs team brings RFC 6265bis compliance with stricter validation and safer defaults. A small but solid ESM-native cookie parser for anyone building HTTP infrastructure in TypeScript.

pi-interactive-subagents — @DanielGri's extension for spawning non-blocking sub-agents in terminal multiplexer panes. Call subagent() and it returns immediately while the sub-agent works in its own pane. Results steer back into the main session as async notifications. A live widget shows all running agents with elapsed time and progress.

@yazins's OpenOats is a macOS meeting companion that transcribes both sides of your calls in real-time and surfaces relevant talking points from your own notes. Runs entirely on-device with local speech recognition — no audio leaves your Mac. Pair it with Ollama for fully offline LLM suggestions, or connect to cloud models via OpenRouter. Recent updates added speaker diarization for up to 10 participants, slashed CPU usage from 84% to 17%, and introduced background mode with menu bar integration.

Swarm is a new tool from @penberg for coordinating coding agents in parallel instead of treating them like terminal sessions. The pitch is that parallel agents can offset the latency and drift that show up when a single agent gets stuck or goes off track.

@outsource_ shares a high-performing jailbreak and MLX quant release for Gemma 4 31B on Apple Silicon. Mostly interesting as a snapshot of the still-accelerating model-jailbreak ecosystem rather than something I'd treat as a product recommendation.

@andrewfarah open sourced a CLI for syncing X bookmarks into a local, agent-readable store. The useful bit is not just backup — it makes bookmarks queryable and usable in downstream automation and note-processing pipelines.

@realalexniebuhr shipped an opinionated Astro starter for deploying to Cloudflare Workers. Handy if you want a batteries-included path into the Astro + Cloudflare stack without burning time on setup and wiring.

@criccomini points to Web Quality Skills, a repo of agent-oriented guidance for Lighthouse and Core Web Vitals work. Useful as a reusable review layer for performance, accessibility, SEO, and general frontend quality.

@karpathy makes the case for publishing 'idea files' instead of polished apps when agents can handle most of the implementation details. It's a good articulation of how software artifacts change when turning ideas into working code gets dramatically cheaper.

Ducklings brings DuckDB Wasm to browsers and Cloudflare Workers, and the latest update adds Iceberg catalog support via R2 Data Catalog. That makes it more interesting as a lightweight analytics layer for serverless and edge environments.

@0xSero interviewed the creator of Pi, the coding agent at the core of OpenClaw. Worth bookmarking less as product news and more as a conversation about how agent tooling is being built by the people closest to it.