Hacker News

Article URL: https://github.com/Mantasaur/AI-referral-detector

Comments URL: https://news.ycombinator.com/item?id=47231184

Points: 2

# Comments: 0

Article URL: https://www.technologyreview.com/2026/03/03/1133848/this-startup-claims-it-can-stop-lightning-and-prevent-catastrophic-wildfires/

Comments URL: https://news.ycombinator.com/item?id=47231182

Points: 1

# Comments: 0

I’ve been using Claude Code pretty heavily lately (plan mode, multi-agent runs, long tool streaks), and I noticed a pattern:

When the AI is busy running tools, I’m usually just… sitting there.

Forty-five minutes later I realize I haven’t moved at all.

So I built a small side project called Claude Gym.

It watches Claude Code’s local JSONL logs and uses them as a signal for “AI is busy vs AI needs input.” The JSONL logs expose structured events (tool calls, agent transitions, turn boundaries), which makes them a surprisingly clean signal for detecting human idle windows.

When Claude doesn’t need me, it pops up short movement prompts (squats, stretches, wall sits). When Claude needs input again, it gets out of the way. There’s also a small pixel-art cat that jumps when a turn finishes.

No network, no telemetry — it just tails the local logs. Written in Go, runs in a separate terminal tab.

It’s intentionally small and a bit playful, but it’s been surprisingly effective at getting me to move during long sessions.

Repo (MIT): https://github.com/477-Studio/claude-gym

Curious whether others have experimented with tying human breaks to agent activity instead of timers.

Comments URL: https://news.ycombinator.com/item?id=47231176

Points: 1

# Comments: 0

Article URL: https://www.stainless.com/blog/sdk-code-mode

Comments URL: https://news.ycombinator.com/item?id=47231175

Points: 1

# Comments: 0

Article URL: https://vibesprings.net

Comments URL: https://news.ycombinator.com/item?id=47231172

Points: 1

# Comments: 0

Article URL: https://github.com/yeominux/md-feedback

Comments URL: https://news.ycombinator.com/item?id=47231167

Points: 1

# Comments: 0

Article URL: https://blog.lyc8503.net/en/post/llm-classifier/

Comments URL: https://news.ycombinator.com/item?id=47230871

Points: 1

# Comments: 0

Article URL: https://www.kelleycom.com/blood-moon/

Comments URL: https://news.ycombinator.com/item?id=47230861

Points: 2

# Comments: 0

Article URL: https://www.cnbc.com/2026/03/03/openai-sam-altman-pentagon-deal-amended-surveillance-limits.html

Comments URL: https://news.ycombinator.com/item?id=47230855

Points: 2

# Comments: 0

Article URL: https://erthalion.info/2026/02/03/new-linux-for-postgresql/

Comments URL: https://news.ycombinator.com/item?id=47230847

Points: 1

# Comments: 0

Article URL: https://cto.new/blog/we-built-an-ai-sre-agent-in-two-days

Comments URL: https://news.ycombinator.com/item?id=47230845

Points: 1

# Comments: 0

Hi HN,

Like many of you, I've been tracking the "Sora-level" video models coming out of ByteDance and Kuaishou. While the tech is impressive, the accessibility often sucks—many are locked behind mobile apps (like Higgsfield) or fragmented invite-only portals.

I built SeeVideo to provide a unified, web-first workspace specifically for Seedance 2.0 and Kling 3.0.

Why I built this:

The Web Gap: Many users are searching for "Higgsfield Online" only to find it's mobile-only. I wanted a desktop-class environment for professional prompt engineering.

Model Nuances: Seedance 2.0 (ByteDance) handles complex physics like liquid and hair differently than Kling. I’ve included a "Prompt Transformer" to help adapt generic prompts into the specific "dialects" these models prefer.

Side-by-Side Benchmarking: It’s hard to judge "cinematic quality" without a direct comparison. This tool lets you see how they stack up on the same prompt.

Technical Observations: From running ~1000 generations, Seedance 2.0 seems to have a higher "motion ceiling" but requires much more specific lighting descriptors compared to Kling's more "forgiving" default style.

I’d love to get the community's feedback on the UI and hear about your experience with the temporal consistency of these latest Chinese diffusion models.

Check it out here: https://seevideo.dance/

Comments URL: https://news.ycombinator.com/item?id=47230843

Points: 1

# Comments: 0

Hi HN, I built Cloudstic (https://github.com/Cloudstic/cli), an open-source Go CLI that creates encrypted, deduplicated, point-in-time backups of your Google Drive, OneDrive, and local files.

The Problem Google Drive and OneDrive are sync tools, not backup tools. If you accidentally delete a folder, it’s gone everywhere. If ransomware encrypts your local files, the "mistake" propagates to the cloud immediately. There is no easy way to say: "Restore my entire Drive exactly as it looked on January 15th."

Why not just use Restic or Borg? I’m a huge fan of Restic, and Cloudstic’s content-addressed design is heavily inspired by it. However, tools built for local filesystems struggle with cloud semantics: 1. File Identity: In Google Drive, files have unique IDs and can have multiple parent folders. They don't have a single canonical path. 2. Efficiency: Scanning a 1TB Drive to find changes is slow and hits API rate limits. 3. Structure: Restic’s tree structure mirrors a directory hierarchy. Cloudstic uses a Merkle-HAMT (Hash Array Mapped Trie) keyed by file ID instead. This allows us to handle cloud-native metadata natively while maintaining structural sharing between snapshots.

Key Features * Encryption by Default: AES-256-GCM (via standard Go crypto libraries). You can recover via password or a BIP39 recovery phrase. Your storage provider (S3, B2, SFTP) sees only encrypted blobs. * Cross-Source Deduplication: We use FastCDC (Content-Defined Chunking). If the same 50MB PDF exists on your laptop and your Google Drive, it is stored exactly once in your backup repository. * Incremental via APIs: Instead of walking the whole tree, we use the Google Drive Changes API and OneDrive Delta API. It only fetches what has changed since your last snapshot. * Crash-Safe & Immutable: All objects in the repository are append-only. An interrupted backup or a network drop leaves zero corruption. * Performance: In my benchmarks (1GB dataset / 10k files), Cloudstic edges out Restic on initial local backups (1.2s vs 1.8s) and matches it on S3 uploads while maintaining a small, bounded memory footprint.

Quick Start Cloudstic is a single static binary. You can point it at local disk, S3/R2/MinIO, Backblaze B2, or SFTP.

brew install cloudstic/tap/cloudstic cloudstic init --backend s3 --bucket my-backups cloudstic backup --source gdrive cloudstic list-snapshots cloudstic restore --to ./recovery

I’m also working on a managed version (https://cloudstic.com) for those who want the same engine but don't want to manage their own crontabs or S3 buckets. The code is MIT Licensed. You can find the full storage spec and encryption design doc in the /docs folder of the repo. I’d love feedback on the HAMT implementation and the crypto design!

Comments URL: https://news.ycombinator.com/item?id=47230835

Points: 1

# Comments: 0

Article URL: https://www.wired.com/story/jolla-phone-2026/

Comments URL: https://news.ycombinator.com/item?id=47230834

Points: 1

# Comments: 0

Hey HN,

I built VibeWhisper because I was using Claude Code all day and the bottleneck wasn't thinking — it was typing. I tried an existing voice-to-text tool, loved the workflow, then saw the price: $20/month for a thin wrapper around OpenAI Whisper where 10 minutes of API usage costs $0.06. So I built my own in a weekend.

How it works: Hold a configurable global hotkey (default: Right Command), speak, release. Transcribed text appears in whatever text field has focus — VS Code, Terminal, Slack, browser, anything. No clipboard hijacking — it uses the macOS Accessibility API to inject text directly at the cursor position (same APIs screen readers use).

Two transcription engines:

Cloud: OpenAI Whisper API via your own key (~$0.006/min, fastest and most accurate)

Local: WhisperKit running on-device via CoreML/Apple Neural Engine. 10 models from Tiny (50 MB) to Large V3 Turbo (1.6 GB). Your voice never leaves your Mac. No API key needed, no internet needed. The app auto-recommends the best local model for your hardware.

Tech details for the curious:

Native Swift/SwiftUI, no Electron. Uses less RAM than a Chrome tab.

CGEvent taps for global keyboard hooks

Accessibility API for text injection

API key stored in macOS Keychain

Universal binary (arm64 + x86_64), macOS 14+

No backend, no accounts, no telemetry

Pricing: $19 one-time. No subscription. 15-minute free trial (actual recording time, not calendar time), no credit card, no signup. If you use cloud mode, you pay OpenAI directly at their rate. If you use local mode, it costs nothing after purchase.

I use it myself 2-3 hours a day. My main use case is dictating long prompts to AI coding assistants — I can speak a 400-word prompt in 45 seconds instead of typing it in 3-4 minutes. I also use it as a thinking tool: talking through problems out loud and capturing the output.

Website: https://vibewhisper.dev

Happy to answer any technical questions.

Comments URL: https://news.ycombinator.com/item?id=47230833

Points: 1

# Comments: 0

I built TubeNitro because I found YouTube’s 2x long-press feature both useful and frustrating.

It’s great for quickly skimming, but 2x is often too fast for tutorials and too slow for boring segments. I wanted something fluid — more like scrubbing with a mouse — but for playback speed.

TubeNitro lets you long-press anywhere on the video and slide your mouse vertically: • Slide up → accelerate (up to 10x) • Slide down → slow down (down to 0.5x)

The speed changes continuously and shows a small overlay so you can see the exact rate.

Comments URL: https://news.ycombinator.com/item?id=47230824

Points: 1

# Comments: 0

I was listening to a recent YC podcast about "build something agents want", and while building apps with Claude Code, I kept running into trouble figuring out how APIs actually behave in practice. And knowing that most companies won't update their docs to be agent-friendly, it clicked — agents need their own Stack Overflow.

AI agents are consuming APIs at a scale and speed that human-written documentation can't keep up with. AgentThreads is a community directory where agents discover, submit, and review APIs — for agents, by agents.

The idea came from watching agents fail silently on undocumented edge cases, relying only on Google searches and bad docs. They need structured, community-maintained knowledge they can query at runtime.

A few things we built: - API directory with agent-written reviews and ratings - No UI, just REST — built for agents, not humans - Anti-spam verification (agents solve reasoning challenges, not CAPTCHAs) - Karma system so agents build reputation over time

Built mostly with Claude Code over a few weekends.

Early days — would love feedback from people building agents on what's missing.

Comments URL: https://news.ycombinator.com/item?id=47230810

Points: 1

# Comments: 0

Article URL: https://hicks.design/journal/my-perfect-music-app-doesnt-exist

Comments URL: https://news.ycombinator.com/item?id=47230804

Points: 1

# Comments: 0

Article URL: https://www.ft.com/content/653fabd7-03da-467a-b2bf-03f226fe2a29

Comments URL: https://news.ycombinator.com/item?id=47230791

Points: 1

# Comments: 0

The SaaS growth playbook has changed: retention now beats acquisition 2.5x. AI-powered companies grow 27% faster and reach $100M ARR 1.7 years sooner.

Key shifts: - Median ARR growth: 35% (2021) → 15% (2026) - CAC payback: 18mo → 23mo; cost per $1 ARR: $1.24 → $2.08 - NRR drives growth: >106% = 2.5x faster than peers - Expansion ARR = 40% of total new (50%+ at scale)

AI transforms retention: 1. Product: predictive analytics, churn-preventing workflows 2. Operations: agents automate CRM, save reps 8-12 hrs/week

Data quality crisis: manual CRM misses 20-30% of critical events, distorting NRR.

Implementation: - Phase 1: clean CRM data, set baselines - Phase 2: email automation, lead scoring - Phase 3: CRM agents, automated NRR - Phase 4: scale personalization

Real examples (our portfolio): - xbeast.io: 3-5x engagement, 2-3x follower growth - nextblog.ai: 35.8% organic traffic growth - reddbot.ai: Reddit-sourced churn 25% lower - vidmachine.ai: 30% higher activation via video

Quantified: $5M→$10M takes 2.3 years (AI) vs 4.7 years (traditional) — a $75M valuation advantage at 15x multiple.

Pitfalls: start with problem not tool; fix data first; augment humans, don't replace; focus on NRR/GRR not vanity metrics.

---

Building AI tools for SaaS growth? Follow @0xWuki for trenches insights.

*Disclosure*: Our portfolio: xbeast.io, nextblog.ai, reddbot.ai, vidmachine.ai. Data from real customers & benchmarks.

Comments URL: https://news.ycombinator.com/item?id=47230780

Points: 1

# Comments: 0