News for hackers. Stuff that matters.

miguelrios

Re: on Friday November 13, 2026 @01:48AM (#50479332)
Yikes. I missed a word or two there... I meant workflows using parallel's web-grounded decisions API. It's like Jev but grounded with their web search https://docs.parallel.ai/decisions/decisions-quickstart.

You can read more about how we turn agents into workflows here and examples of using Parallel's decisions API here: https://agentrun.ai/blog/you-may-not-need-an-agent-after-all

Re: on Friday November 13, 2026 @01:01AM (#50478954)
You can create the agents inside of Agentrun and pick what harness / model to use. You can also install them from templates. https://github.com/Parcha-ai/agentrun#templates

I am publishing a library of templates tomorrow with many more.

Re: on Thursday November 12, 2026 @10:12PM (#50477696)
uh, interesting. I will take a look into this.
Re: on Thursday November 12, 2026 @09:25PM (#50477294)
It supports Claude Agents' harness but their sub is still not supported. I am working on it. Same for Codex / ChatGPT.
Re: on Thursday November 12, 2026 @07:46PM (#50476642)
This is fair feedback. We spent a ton of time in the UI for our hosted product and the main website (agentrun.ai) but very little in the OSS console. We’re spending more time on this now and making it better.
Re: on Thursday November 12, 2026 @07:45PM (#50476626)
Yes here is an article about it but I’ll will explain the tldr below. https://agentrun.ai/blog/agentrun-a-harness-for-repetitive-k...

The core concept is to have a frontier or very smart model run as an agent first for N times. The model get tools, skills etc and instructions. It does the job. At the end it leaves a note (like a retro) explaining what it did and what it could have done better.

As the agent runs, it also writes workflows to do the parts of the job it knows it can do in a more deterministic way (based on previous nodes and thinking traces of other runs etc). We made a dsl for this that has routing, classification, code nodes, etc to build these workflows.

For repetitive tasks (which is our focus) we find that after ~100 cases, the agent most times just runs a workflow end to end instead of looping towards a solution. Workflows are cheap, so the job ends up being way cheaper than an agent in a good model.

Re: on Thursday November 12, 2026 @05:19PM (#50475160)
very cool. thanks for sharing!
Re: on Thursday November 12, 2026 @05:03PM (#50474965)
Would love to hear that too... it's getting incredibly expensive. One of our templates is a code review agent. https://agentrun.ai/blog/saving-80-percent-on-code-reviews-u.... But we're looking to do more beyond making the review cheap.
Re: on Thursday November 12, 2026 @04:03PM (#50474243)
Do you have an MCP? I am adding our internal tool proxy which can be used to plug in any MCP into it.
Re: on Thursday November 12, 2026 @02:56PM (#50473271)
Very cool. Feel free to borrow from us what's helpful and send feedback our way!

Show HN: AgentRun – open-source, Self-improving Managed Agents

https://agentrun.ai/blog/introducing-agentrun-run-smarter-agents-use-fewer-tokens

We’re open-sourcing AgentRun and turning it into a full managed agents platform. It includes what we originally built for our own agent harness:- Self-improving agents - Web-grounded workflows - A range of sandboxes and computer environments Thousands of tools and integrations - Detailed analytics on how your agents behaveYou can bring your own agent. AgentRun works with Pi, Claude Agents,...

( Read More... | no comments | on HN )

Show HN: AgentRun: DSL to turn agents into Workflows

https://github.com/Parcha-ai/agentrun

Hi HN,I just open sourced the DSL that our harness in grep.ai uses to turn repeatable parts of agent work into workflows. You can combine tool calls, code, Jev-powered system one decisions for things like routing and screening evidence, and agents when a step needs more investigation.Our harness uses the traces and retro notes agents leave behind when doing a job to figure out which parts can...

( Read More... | 11 comments | 8 points | on HN )

Show HN: Start and Up – AI Music Album and Visualizer Produced by Claude Code

https://cen.so/startup/

Like many others, I've been experimenting with using Claude Code for non-coding tasks. The coding portion of this one was a small percentage of the project. All songs are generated by Suno, but each one was produced by Claude Code using an iterative process that ended with annotations, lyrics, and style.In the site, you can read an explainer of each song, the album, and even an analysis of the...

( Read More... | no comments | on HN )
Re:Facebook has the world's largest Hadoop cluster on Tuesday September 06, 2011 @01:40AM (#2964305)
"Sunday, May 9, 2010" just saying... those numbers may be different now.
I am a big fan of these guys. While engineers are underpaid and schools don't promote the startup culture at all in Puerto Rico, they have stayed loyal to hack and build stuff in their own in the island. They deserve at least your vote or advice.
Re:Arbor.js - HTML5 graph visualization library on Wednesday January 12, 2011 @04:02PM (#2097683)
Very interested on this. How scalable it is? How many nodes can I see?

I ask because it uses a Force Directed layout, I guess it will get very slow after a few thousands of nodes.

Looks promising though. I'll play with it...

Re:Mixpanel - Internships. A story. on Tuesday January 04, 2011 @05:07PM (#2068734)
Wow... you just explained my experience at Twitter. I went from being a future PhD looking for a fun summer to someone who forgot about grad school and wanted to stay creating "real" stuff in 4th and Folsom for ever.

Suffice it to say, I ended my internship, left grad school (including a fellowship by Google via GEM) and stayed at Twitter.

Get an internship, it can change (or confirm) your vision about a career and your future in general.

Re:Poll: Is HN your main online news source? on Wednesday December 29, 2010 @01:07PM (#2049585)
For me it is a mix os HN, Reddit and Techmeme, all consumed either from Twitter (any time) or from the occasional browsing (three - four times a day).
Re:Facebook intern visualizes friendships, draws world map on Tuesday December 14, 2010 @02:13AM (#2003426)
I bet you Gephi cannot handle more than tens of thousands of node.
Re:Hacker News firehose feed on Saturday November 13, 2010 @01:09PM (#1901264)
Would be awesome if you add the link to the HN page too. Sometimes I want to read the comments too.
Re:⌘Q is too damn close to ⌘W on Monday November 08, 2010 @11:14PM (#1885294)
That doesn't change the fact that I get mad and lose time whenever I accidentally close all my tabs.