Hacker News

New stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. How we share, review, and update AI agent skills(medium.com/ali.raza_25708)
    discuss
  2. Calibrating Jev as a Code Reviewer(github.com/selmar)
    discuss
  3. Firecrawl Alexandria(firecrawl.dev)
    discuss
  4. Agent Incident Registry(enkryptai.com)
    discuss
  5. When your Mac just won't let go of the Wi-Fi connection(sixcolors.com)
    discuss
  6. The First Microbe for Mars(pioneer-labs.org)
    discuss
  7. The AI Party House Where Networking Has a Dark Side(nytimes.com)
    discuss
  8. Wiring Diagrams for BMW Radios(bmw-radios.com)
    discuss
  9. Show HN: Hindsight: Automated Work Timeline(apps.apple.com)
    discuss
  10. Study Sessions Should Be Short and Frequent as Opposed to Long and Sparse(justinmath.com)
    discuss
  11. Show HN: AI·rete·RAG – a Rete rule engine decides, RAG explains why(ai-rete-rag.com)
    discuss
  12. Call of Duty's Activision to make next Halo game as Xbox cuts more jobs(bbc.co.uk)
    discuss
  13. Gravity-Gradient Stabilization(wikipedia.org)
    discuss
  14. Satsie's Pocket Guide to Op_cat (Bitcoin)(satsie.dev)
    discuss
  15. 16-bit Intel 8088 chip by Charles Bukowski(allpoetry.com)
    2comments
  16. Xbox continues its "reset" with dramatic restructuring(arstechnica.com)
    discuss
  17. 6 Years In, $6M Far(tally.so)
    discuss
  18. Jev-blindspot – find the blind spots in the prompt you just sent(github.com/jsk4581)
    discuss
  19. Show HN: Every Credit Card(numberplanet.com)
    1comments
  20. Research Harness from First Principles(edotenv.com)
    1comments
  21. Tell HN: Firefox's default browser check dark pattern triggers Windows BSOD
    1comments
  22. Streamhouse: What It Is, What It Isn't, and What's Missing(conduktor.io)
    discuss
  23. No More Warning Shots(aenguslynch.com)
    discuss
  24. Raspberry Pi locks boards to factory RAM capacities in firmware(tomshardware.com)
    discuss
  25. 1,300 STEM 3D Models created by GPT-6 Astra(emergentmind.com)
    discuss
  26. The Home Computer Generation(datagubbe.se)
    discuss
  27. We built muse from scratch, but it is definitely inspired by OpenClaw(twitter.com/natfriedman)
    discuss
  28. Marvin Heemeyer(wikipedia.org)
    discuss
  29. Our Quantum Future is a documentary exploring the rise of quantum technologies(ourquantumfuture.com)
    discuss
  30. Founding Teams(ajays.quest)
    discuss

Show HN: Kado – Search to Allow Personal Agents to Find Agent Apps

2 pointsby 48m agokado.so
0 comments
Hey folks!

My cofounder and I have built Kado - Search for Personal Agents to find Agent Apps[1].

We think the future of agents that individuals will use few personal agents (think, personal computers) as their primary interface, and while those agents delegate work to agent apps to complete actual tasks.

We were building a personal agent ourselves when we noticed that our users were often trying to do complex things which should be delegated away. We found that web search didn't cut it for reasons including the "quality metric" being incorrect, and a "don't know what to search for" problem.

Kado's crawl uses LLMs to parse information from websites pages, agent cards, or other discovery sources to model the actual outcome(s) that's being promised, and the interface to them (we call these "agent apps").

Our search and index is based on a heavily modified version of HyDE. The search itself is quite slow right now because of LLMs in the loop. We want to get this down to ~sub-second.

We have ~3000 indexed MCPs and/or A2A agents, and we recently purged the index of entries your agents can't use (we had ~45k before that). I'm partial to a TTRPG character generator MCP that's great for printed one shot sheets, but some others you may also want to try searching for include builder/research/crawl tools or (surprisingly) financial data tools. Also, did you know that AWS has a public devops/root cause analysis agent?[2]

The best way to try it is to install it on your agent (it's a CLI + Skill for now, MCP coming soon!) but you can also just try direct searches on the website. The website requires a google login for spam protection (which we want to remove if we get some funding haha), but the CLI does not need a login (we use proof of work to prevent spam, and an agent identity based on a keypair is auto-created and stored in your keychain).

Would love any feedback!

PS: Ignore the order of the search results - our experiments show that the ranking does not matter for intelligent models, so we've prioritized recall over ranking.

[1] https://kado.so and https://kado.so/about [2] AWS DevOps Agent - https://connect.aidevops.us-east-1.api.aws/.well-known/agent...

A quiet thread, for now.Start the conversation on HN ↗