
tl;dr: OpenRouter became the router for model spend; we're building that layer for agent capabilities, starting with paid tools: web search, scraping, extraction, enrichment, and more from top providers in each category. No provider accounts or API keys, one balance, pay per use. We benchmark every provider on real tasks and route each request to the one that performs best, with fallbacks when a provider fails. We're looking for early users to try it out with free credits, email us at founders@amorphiclabs.com if you want to check it out!
Launch Video:
The problem:
Agents can only use the tools a human wired up in advance, and they won't tell you what they're missing: they quietly make do, failing at anything they weren't hand-equipped for. Even when you do the wiring, you're picking blind: nobody actually knows which search API is better for this query, or which scraper will survive this site. The "best setup" frontier moves daily, so keeping up means hours of trawling MCP directories, X threads, and blog posts. Agents stay only as capable as their setup, and the more autonomous they get, the more that ceiling costs.
What we do:
Your agent asks for what it needs ("structured data from this page", "research this with citations"). We route the request to the best provider for that specific task, based on benchmarks on real tasks and outcome feedback from the agents using them: success rates, result quality, latency, and cost. If a provider fails, we cascade to the next, so your agent gets a result instead of an error. Every request makes the routing smarter. One balance pays for all of it, and it's the same MCP whether it's your own coding agent or agents you're shipping in a product.
APIs are just the start. The same marketplace will let agents acquire skills, hire other agents for specialized tasks, and sign up for software directly through MCPs, all from one balance.
Our team:
We're Frank and Dylan, friends of 10 years who have been building together since middle school. Frank was a Member of Technical Staff at OpenAI and interned at Two Sigma, NVIDIA, and Amazon. Dylan was a quant trading intern at UBS, co-authored a NeurIPS paper, and was a grand finalist and 8th-best speaker in the world at WSDC 2022. We built this because we kept hitting the problem ourselves: agents stuck mid-task waiting for us to make accounts and guess at providers.
Our ask:
Make something agents want!