Introducing Offloop! We're a team of four. Today our multi-agent harness hit state of the art on GDPval, ahead of Claude code and Codex across jobs that pay $2.4 trillion a year in the US. Offloop gives every knowledge worker what the Fortune 500 spends billions on: a high-performing... agent army that runs itself and grows the business.show more

Here is how Alice runs an entire company inside Offloop. Alice is building a private, local-first legal AI. She started with one instruction. Offloop assembled the agents army, organized their work into channels, and set up a shared flow across product and engineering team. The agents build in parallel - the desktop runtime, Word revisions grounded in verifiable citations, client documents locked to the device. When QA caught a restart failure, engineering fixed it while testing kept running. Alice reviewed the evidence in one thread and approved a limited private beta. Public release stays gated. One instruction in. One approval out. Everything in between was agents.

Meet D1, the world's first dispatcher model, and the best way to orchestrate a team of agents. When a client email lands in your workspace, something has to decide what happens next - which agent takes it, who reviews the draft, and which six agents should stay out of the way. That's D1's entire job. It's trained on real production dispatch decisions, and it's why our agents finish work without talking over each other or burning tokens they didn't need. On Dispatch Bench, our eval built from those real interactions: D1 scores 81%.

Offloop is still early. We are learning what context should last, when agents should act, and where human judgment matters most.

join the waitlist:

@sama bet we’d see the first one-person billion-dollar company. We think the ceiling is at least 10x higher. Every David should have a real chance to defeat a Goliath.

@sama a big day for agents!

Nice!! Shared context across agents is the unlock. Tired of re-explaining my project to every new chat

LFG

Congrats on the launch! Looks like a great way to work with agents in teams. How does D1 decide when to pull a human in versus letting the agents keep going?

Great job! Congrats on the launch! We need a better way to communicate with agents. Mentions on Slack are terrible idea.

big day for agents

agent interface moves to a new milestone - I’m not joking

shipping something this capable with a team of four is crazy

Congratulations guys, this is sick - can I scale up my dropshipping business?

One of our early hypotheses is that AI-native commerce teams will be among the biggest winners!!! Would love to put that to the test with you🙌🙌

Big vision. Looking forward to seeing how it evolves. 👏

Here's how we run our entire company on Offloop. Across marketing, sales, and fund raising

Is GDPval the right benchmark or just the best one available today?

It's the most relevant benchmark we've found so far with a reasonable level of credibility. Our target use case is actually a subset of what GDPval evaluates. We also believe the ecosystem will eventually develop much better benchmarks specifically for measuring multi-agent collaboration, and we're looking forward to that.

Running entire company using Offloop is wild

What happens when two agents disagree on an edit?

Sometimes a third agent joins the discussion and makes the final call.

Congrats on the launch, this is a big step forward! 🎉👍

Small teams are proving that agent orchestration is becoming the real competitive advantage.

this is impressive 🫡

5X cheaper damn....

We've been looking for a solution like this for our team. Could you spare an invite code so we can put it through some real workflow tests? Thanks!

I spend too much time re-sharing context between AI tools. This feels like a much better approach.

This is such an underappreciated concept, yet critical for managing costs and outcomes. Love to see the innovation! 🦾

I was thinking about fine-tuning a slm as a router but you executed 10x better than I could do! Well done!

How do I get an invite code?

Sure, please check your DM

Congratulations on the launch

Multi-agent systems are opening exciting possibilities for the future of knowledge work.

Curious to see how Offloop performs outside benchmarks. Real production workflows will be the real test.

This seems pretty cool! Can't wait to try it! Great job!

Congratulations to the Offloop team on reaching state-of-the-art performance.

GDPval is a solid benchmark. Curious how you handle state consistency across those agents during long-running tasks.

congrats on the launch The interesting part is not only benchmark performance but also the idea of agent orchestration.

Excellent comparison, very insightful.

Congratulations to the fantastic 4

Congratulations! Excited to see autonomous agent teams tackling real-world knowledge work at scale.

Go on. •

Can I get invite codes for my team. Growth team of 4 people. Would love to try it out and probably implement in our workflow.