
ClawsGO Science
The AI research agent with its own cloud lab
416 followers
The AI research agent with its own cloud lab
416 followers
Research, fully automated. ClawsGO Science takes a question to discovery and writes it up as a cited paper. Its cloud lab has 500+ tools and 90+ datasets. Its long-horizon harness runs clusters of Claude and GPT agents for days. The next paradigm for science.






Free Options
Launch Team

AWS Activate CreditsCredits that fuel your startup’s next milestone
Promoted
Congrats!
@arthur_winston3 Thank you! Means a lot on launch day. We're here all day if you have questions.
A cited paper at the end instead of a chat answer is a strong output format. How can researchers check which tool or dataset each claim came from?
@suleymankasikci Great question, and it's the one that matters most in science. Every search, command and file edit is on record. Open the run and you see which tool ran, on which data, and what it returned. Replay the whole run, or share it with a public link so a colleague can check it too.
Package setup and LaTeX cleanup are usually where I lose time. Curious how much of that ClawsGO handles on its own.
@jayzhu Most of both. Packages: 500+ tools are preinstalled, and if one is missing, the agent installs it on the workstation itself. LaTeX: the agent writes the source, compiles it in the cloud, reads the log and fixes errors until the PDF builds. You get source and PDF side by side to edit.
The preloaded lab is the part that caught my eye. Most agents I've tried stall on setup before doing anything. How much setup does a typical first run still need?
@sarasolomaker For most first runs, none. Sign up, type your question, and the agent starts on a workstation where the tools and reference data are already installed. The only thing you might add is your own data. Drop the files into the task. Ready on minute one.
the self-review step is the part I keep thinking about - rereading a draft against sources and code will catch a citation mismatch, but what about when the agent just misread what the data actually shows in the first place? that feels like a harder failure mode than a wrong citation. does self-review catch that too, or is that still on the human to catch?
@galdayan Great question, and you're right, it's the harder failure. A citation check can't catch a wrong reading of the data. What helps: every step is on record, so you can open the code and output behind any claim. You can also ask for checks up front: plot the raw data, run sanity checks, reproduce a known result first. For anything you publish, the final read is still yours.
@annactswell "reproduce a known result first" only works when you know what result to reproduce. for a genuinely novel question there's nothing to anchor against yet - no ground truth to check the pipeline against before it goes off and runs the real analysis. is a mandatory sanity-check step something the harness forces before novel work starts, or does the user have to remember to ask for it?
The citation check is the feature I would test first. Can users see exactly how each reference was verified?
@yiyaoliu You can trace it in the run record. Every search and every page the agent opened is logged, so each reference leads back to where it was found. Want a formal check? Ask it to verify every DOI against Crossref and PubMed, and list the result for each reference.
running Claude and GPT agents side by side for days is a wild setup. what happens when they come back with different answers to the same question?