
From the team
Blog
What we ship, how it performs,
and practical ways to put rtrvr to work.


Product
LinkedIn and X in Cloud: Tailscale, saved logins, and cookie sync
Find customers, comment on posts, and send personal messages from Retriever Cloud. Connect your accounts, keep sessions fresh, and use your own IP with Tailscale.
Read article
Product
Break the lethal trifecta: sandboxed cloud agents on your own network
An agent that holds private data, reads untrusted pages, and can send data out can be turned against you. Here’s how we scope each cloud agent’s logins, network, and files, and how its traffic exits through your own Tailscale node, plus the DNS rebinding gap still open.
Read article
Product
For you: AI suggests what to automate
Discover workflows you can hand off to Retriever AI, based on your recent activity and the browsing context you choose to share.
Read article
Research
What People Ask AI Agents to Automate
What 46,476 requests from 1,023 rtrvr users reveal about the work people want AI agents to handle.
Read article
Proof
OpenAI dots vs rtrvr.ai: Speed, Cost and Accuracy
rtrvr ran five tasks about 5× faster than OpenAI dots. See the results, invoice recovery and browser-agent harness, then try the tests yourself.
Read article
Research
How OpenAI Agents Hacked Hugging Face to Cheat a Test
How OpenAI agents coordinated to cheat a cybersecurity test, hacked Hugging Face, and used a screenshot service to run code. Evidence and browser-agent defenses.
Read article
Proof
Muse for everyday life. Retriever AI for work.
Muse missed earlier instructions and called visible Instagram likes hidden. Why context management and on-device browser access matter for creator research.
Read article
Research
How Jev Chooses the Next Browser Action
We added Jev to rtrvr’s browser agent to choose from the actions available on a page. Our LinkedIn and Amazon runs finished sooner, but scoring large pages raised the cost.
Read article
Research
Stop Paying AI to Plan the Same Task Twice
In a new paper, a 9B model with tested skills beat a 27B model without them. Browser agents now need a way to find the right skill, test it, and remove it when it stops working.
Read articlePUT IT TO WORK