Every agent hits the same wall: the web. Scraping breaks, MCP burns tokens, logins kill automation. TinyFish, trending on Product Hunt today, sells the fix as infrastructure — a web operating layer for AI agents, backed by a $47M ICONIQ-led Series A.
One API key, four primitives
Web Agent, Search, Browser, and Fetch under a single key and credit pool. Their benchmark: a CLI-style call costs ~100 tokens versus ~1,500 over MCP — an 87% cut — with task completion roughly doubled. Search and Fetch are free. Typical use case: point the Agent API at any site in natural language, get structured live data back — Google uses it to aggregate thousands of Japanese hotel inventories for hotel search.
Mako is the real moat
In July TinyFish shipped Mako, a 35B MoE model (3B active) post-trained on a year of production enterprise web tasks — authenticated, multi-step, real consequences. Browser-use startups mostly wrap frontier models; TinyFish trained its own on data nobody else has. It also open-sourced BigSet, a multi-agent system that turns natural language into structured live datasets. Infrastructure plus a proprietary model plus enterprise logos is a much harder position to attack than another browser agent demo.
You Might Also Like
- Insforge Hits 1 on Product Hunt and 3600 Github Stars is This What Agent Native Backends Look Like
- Design Agent by Lokuma Topped Product Hunt With a Simple Pitch ai Agents Need Designers too
- Google Search Information Agents Turn 1 Billion ai Mode Users Into Agent Operators
- Opencomputer Agent Deploy Turns one Prompt Into a Live Agent url hit 4 on Product Hunt
- Google A2ui Agent to User Interface Finally a Standard way for ai Agents to Show you Things

Leave a comment