Add to Chrome
From the team
Blog
What we ship, how it performs,
and practical ways to put rtrvr to work.
Research
Financial Markets and Model Routers
Stock markets aggregate information for price discovery. Blockchains distribute execution and settlement. Model routing can combine both architectures: centralized data for selection, local provider estimates, and independent verification of task completion before updating the next route.
Read article
Research
How We Built a Free Browser Agent, Paid for by Ads
How rtrvr cut a browser-agent task to about half a cent using semantic page trees, code plans, and prompt caching, then made the agent free with sponsored cards.
Read article
Research
How prefix caching cut our browser-agent token bill by 90%
The cache boundary, measurements, and implementation choices that made long browser runs materially cheaper.
Read article
Product
The self-improving action layer we are building
A concrete roadmap for turning successful browser work into reusable actions without losing review and control.
Read article
Proof
From 500 event attendees to 10 prospects
How a browser-based enrichment workflow researched an attendee list, kept its sources, and prepared the best matches for review.
Read article
Research
Websites need headless agents, not chatbots
Browser agents are a compatibility layer. The durable interface is a site-side agent that exposes intent handoff, knowledge, auth, tools, payments, and policy.
Read article
Research
A cheaper browser agent: let DeepSeek write the plan as code
What changed when we replaced repeated tool selection with a compact program executed against a text-first browser state.
Read article
Research
Navigation Was Never the Moat: Why Agents Make the Sell Side Bigger, Not Smaller
Models will soon navigate any website perfectly. That was never the moat. The agent that converts is the one the seller owns — its accounts, its payments, its data — and the world's biggest retailer just proved it inside ChatGPT.
Read article
Research
Replacing browser-agent tool loops with eval() in a Chrome sandbox
A for-loop should not cost tokens. We let the model write browser workflow code once, then execute it locally through a sandboxed `rtrvr.*` DSL layer.
Read articlePUT IT TO WORK