惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Last Week in AI
Last Week in AI
阮一峰的网络日志
阮一峰的网络日志
P
Proofpoint News Feed
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
MongoDB | Blog
MongoDB | Blog
云风的 BLOG
云风的 BLOG
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
J
Java Code Geeks
WordPress大学
WordPress大学
T
The Blog of Author Tim Ferriss
V
Visual Studio Blog
小众软件
小众软件
Microsoft Azure Blog
Microsoft Azure Blog
博客园_首页
IT之家
IT之家
Vercel News
Vercel News
C
Check Point Blog
Google DeepMind News
Google DeepMind News
月光博客
月光博客
D
DataBreaches.Net
酷 壳 – CoolShell
酷 壳 – CoolShell
美团技术团队
Y
Y Combinator Blog
Hugging Face - Blog
Hugging Face - Blog

Show HN

GitHub - astefanutti/shaderbang: Shebang for Shaders Show HN: Generate Claude Code Workflows using Spec Driven Development approach Show HN: AI agents for UK GDAD PCF roles and their skills The Two Pillars: Mixer Mode and Meta-Software in the Reorganization of Software Work After AI GitHub - JaiCode08/teleport-env What 1,000+ Harness Experiments Taught Me About Self-Improving Agents Show HN: Liiists, a Markdown-first, iOS and CLI list app SwiperTab – Get this Extension for 🦊 Firefox (en-US) GitHub - kouhxp/fftext: Summarize, explain, fact-check, or translate any text, URL, or file. No GPU. No cloud. One command GitHub - sweetpad-dev/sweetpad: Develop Swift/iOS projects using VSCode GitHub - dogmaticdev/IRON: IRON a.k.a. Intermediate Representation Object Notation is a Interpreter/Database that is used to create Programming Languages. GitHub - sjhalani7/vaen: Package your AI coding harness into a portable .agent file, and share it across repos, teams, & the community without ever having to copy-paste instructions, skills, MCP config, or secrets. Show HN: Gandalf the Grader Show HN: Citadeld – replay any CI failure locally from a single file GitHub - tdortman/cuSBF: High-Performance GPU Super Bloom Filter coral-ai/claude-code-token-xray at main · Coral-Bricks-AI/coral-ai GitHub - ulyssestenn/funes: Funes is a Git-based framework for LLM-managed knowledge work: an AI Librarian ingests raw sources, builds an interlinked Markdown knowledge base, and uses it to produce cited reports, analyses, and other outputs. GitHub - ThatXliner/gah: Git Add Hunk, built for agents to use GitHub - harmont-dev/harmont-cli: Command-line client for the Harmont CI platform GitHub - brooksmcmillin/mcp-authflow: OAuth 2.0 Authorization Server framework for MCP servers GitHub - javaid-codes/audit-supply-chain-agents GitHub - amorey/gochan: A small library of common channel architectures for Go, inspired by Rust GitHub - arifozgun/OpenGem: Free, Open-Source AI API Gateway with Gemini, OpenAI & Anthropic Compatibility in 1 file GitHub - Pranesh950/BioPetals: 🌸 Run BIOxAI models at home, BitTorrent-style. Fine-tuning and inference up to 10x faster than offloading GitHub - cnguyen14/bounty-doctor: Diagnose a GitHub bounty issue before you waste hours: detects honeypot scam repos, AI-bot attempt swarms, and stale contests. Show HN: CoreMCP – MCP Server for On-Prem DBs Show HN: KittyHTML – Render HTML/CSS as an inline image in your terminal GitHub - bingud/filemat: Web-based file manager Show HN: TruthLens – Free multi-signal deepfake image detector GitHub - apexlocal-jz/claude-usage-tray: Windows system-tray app showing your Claude Code rate-limit usage at a glance. Zero deps, ~300 lines of PowerShell. Cross-IDE (works regardless of VS Code, Cursor, plain terminal).
ask-a-human: a private pager for agent loops
alexandrosky · 2026-06-23 · via Show HN

ask-a-human

Run your agents full-auto in loops. When one actually needs a human, it pings your phone and waits.

End-to-end encrypted No accounts No database No tracking Open-source

Paste this. That's the whole setup.

No account, no API key, nothing to install or build. Drop this into your mcp.json and your phone is now a tool your agent can call.

mcp.json

{
  "mcpServers": {
    "ask-a-human": {
      "command": "npx",
      "args": ["-y", "@askahuman/mcp", "serve"]
    }
  }
}

Works with your agents

Claude Code Codex Cursor Copilot Gemini Any MCP client

If it speaks MCP, it can reach you. Same one-line config, zero extra setup.

This is what lands on your phone.

Pair once. After that, when an agent can't decide on its own, it rings your phone and waits: a yes or no, a choice, or a quick reply.

Open ask-a-human.ai/app on your phone and type the 10-character code it printed to connect.

ABCDE-23456 type it in the app

Claude Code

stuck on a risky deploy

All tests pass, but this release also runs a migration that drops `orders.coupon_code` on prod. Ship it?

Approve Decline

Codex

blocked, needs your call

`main` has been red for 20 min and 3 deploys are stuck behind it. How do I unblock it?

Revert it Hotfix forward Hold, I'll look

A customer is escalating in the support thread and demanding a refund I'm not allowed to approve. How should I reply?

Type your reply… Send

Wall to phone to “keep going,” in seconds.

1 Agent hits a wall it shouldn't pass alone

2 Sealed in a wormhole

3 Buzzes your phone

4 You tap: approve, decline, or reply

5 Agent unblocks and rolls on

Why we built this

We run a lot of agents. Like a hundred of them, in loops, full-auto, full permissions, mostly while we're asleep or away from the keyboard. That's the whole point: they're not supposed to need us. Until, every so often, one of them does.

And when an agent on full-auto hits the one step it shouldn't take alone, it has no way to reach us. It can't pause a hundred loops and wait by the door. So the choices were both bad: babysit every step and kill the automation, or let it rip and hope nothing touches prod while we sleep. We wanted a third option. Let them all run wide open, and give the rare one that's stuck a way to tap us on the shoulder.

So we built a pager. Copy-paste one MCP server into your agent, pin this site to your iPhone home screen, and that's it. Nothing to install, no account, no API key. When any of your agents needs a yes or no, a choice, or a quick reply, it pings your phone and waits. A hundred agents, one place, one buzz at a time.

And it's private by design. The messages travel through a magic wormhole, end-to-end encrypted, so the relay in the middle is blind. We don't track anything, there's no database, and we pay for the hosting ourselves, all of it. No data to sell, no funnel, no catch. We built it because we needed it, and our only goal is for it to spread.

For the curious

The crypto bit.

Pairing is a Magic-Wormhole-style SPAKE2 handshake: that short code turns into a strong shared key, and even if someone shoulder-surfs it they get exactly one online guess before it's useless. After that, every message is sealed with NaCl secretbox (XSalsa20-Poly1305). The relay only ever sees base64(nonce‖ciphertext) and which room is talking to which, never the contents.

Your agent and your phone derive a shared key from a short code via SPAKE2; the relay in the middle only ever forwards ciphertext. The relay only forwards ciphertext. base64(nonce‖ciphertext) + which room talks to which, never the data sealed sealed Your agent MCP · holds the key Relay content-blind Your phone holds the key SPAKE2: short code → a key the relay never learns

Built so it can't betray you.

Not a privacy policy you have to trust. A design that can't betray you. No accounts to breach, no database to leak, no logs to subpoena. The relay is RAM-only: restart it and you simply re-pair. And if you still don't trust us, the source is right there and you can host your own relay with --relay / --public-relay.

Content-blind relay

The server only ever sees base64(nonce‖ciphertext) and which room talks to which. It's a dumb pipe: never the data, never a decision.

SPAKE2 pairing

A short pairing code becomes a strong shared key via a Magic-Wormhole-style SPAKE2 handshake. No relay-in-the-middle can read or forge it.

End-to-end encrypted

Your agent and your phone hold the keys. Approvals are sealed on one side and opened on the other. Nowhere in between.

No DB, open-source

RAM-only, no accounts, no database. Restart means re-pair. Self-host the relay and web with a flag, or read every line on GitHub.