Remoroo

Overnight, on a Mac, from weak baseline to solved. This is what autonomous experimentation should look like: set a goal, run real experiments, wake up to results. BipedalWalkerHardcore-v3 is just the start.

This is what most robot demos look like before the demo.

Spent twenty years building hard machines. This is what I learned. And what I'm doing now. remoroo.com

Remember when chatbots were a joke? Now every company has one. But I don’t think chat is the real interface. 🧵

Watch @RemorooLabs solve LunarLander-v2 in under 7 minutes.

The hard problem in AI memory is not “can the agent grep the repo?” Of course it can. The hard problem starts after 5,000 tool calls, 30 failed experiments, 12 changed files, 4 baselines, and one metric that moved for the wrong reason. That’s where memory starts to matter.

We asked @RemorooLabs to make our training run faster. It came back with: "Your training run is fast. Your data loader is not." This is what closed-loop AI looks like in practice; fewer hallucinations, less happy-path hand-holding, and a lot less "Great question! Here's how I'd approach that..." Just an agent that finds the actual problem.

We did it again... coming soon

This one is big. Wait for it.

I Gave This AI One Job Before Bed. By Morning, It Had Won.

I let @RemorooLabs run experiments overnight on my Mac, and by morning it had solved BipedalWalkerHardcore-v3. This is a notoriously difficult reinforcement learning benchmark, especially on low-compute hardware like a Mac. Starting from a weak baseline, Remoroo ran autonomous experiments, learned from the results, and found a winning solution. Here’s the video.

Remoroo: Runs overnight experiments on your code locally producthunt.com/products/remor… via @producthunt

Agents building agents! Coming Soon! #autonomousengineering

Most coding agents accumulate context every turn. That is tolerable for short edit loops, but inefficient for long-running research experiments where the agent needs to iterate over hours, not chat history. Remoroo is built for research and engineering loops, not just code generation. remoroo.com/blog/how-remor…

Built Remoroo for long-running engineering experiments. Give it a repo + a measurable goal, and it keeps iterating: patch → run → eval → repeat Try it: pip install remoroo remoroo.com #autoresearch #AutonomousEngineering