Taku

Taku exists because most useful AI setups are locked inside somebody else's repo, config file and API keys. It packages agents and workflows as things you open and run, and puts them in a marketplace you can browse — no cloning, no pip install, no keys to wire up before you find out whether the thing is any good.

Three pieces. The marketplace is a catalogue of AI apps, agents and workflows you can launch as one-click tools. Stacks let you describe a goal in plain language and have Taku pull the matching apps out of the marketplace into a single workflow that belongs to you. Stax is the publishing side: package a workflow you have built, publish it, and other people run it — with a revenue share when they do.

Why it is interesting. The gap it targets is real. There is a large and growing supply of agent recipes — prompts, skills, MCP wiring, multi-step chains — and almost no distribution layer where a non-engineer can try one without a terminal. Taku's answer is a unified API proxy so one account covers the model access several tools would otherwise each demand, and an app-shaped interface so competing setups can be tested side by side rather than swapped in and out of a single chat window.

Scale and price. The site reports 12,000+ apps and skills, 3,400+ creators and around 1.2M runs a month, with nothing to install locally. It is freemium: you can run real workflows on the free tier and upgrade when it becomes part of your work. Taku launched publicly on Product Hunt on 18 August 2026.

Caveats worth stating. A marketplace this open is only as good as its curation, and quality across a catalogue that size will be uneven — treat the top of the marketplace as the signal and the long tail as unvetted. Creator revenue-share terms beyond the basic rev-share are still being filled in.