Introducing Froots — your AI playground

Today we’re opening the doors to Froots — a desktop playground that lets your AI actually do things. Browse the web, take notes, draft code, run shell commands, schedule itself for tomorrow morning. The pieces have existed for a while. What hasn’t existed is a place where they live together, share memory, and stay yours.

Most AI products today force a tradeoff. Chat apps are great at talking but forget you the moment you close the tab. Agent frameworks can drive a browser but ship without a place to live. Note apps want to be your second brain but can’t reason about what they hold. We kept running into the same wall: the interesting work is at the seams, and nobody owns the seams.

What Froots is

Froots is one app with four tightly-linked pieces. Memory is a graph of everything you and your agents have ever said, written, or done — searchable, linkable, and yours. Routines are scheduled agent runs: morning briefings, weekly reports, market checks. Skills are togglable capabilities, more like apps than plugins. Conversations is where you steer it all in real time, with full visibility into tool calls, file edits, and reasoning.

The whole thing runs on whatever model you want — Claude, GPT, Gemini, Llama through Ollama, Mistral through Groq. We’re model-agnostic on purpose. The brain you pick today won’t be the best brain in six months, and we’d rather you swap models than swap apps.

What it isn’t

Froots isn’t a SaaS chat app dressed up as a desktop client. Your data lives on your machine. Your model keys are yours. Your memory graph is a file you can grep. We sync to the cloud when you ask us to, never by default.

How to try it

Open the playground, paste your model key, and ask it to do something small — summarize your inbox, draft a blog post, scrape a website. Watch the run timeline. Pause it, steer it, rerun it. That’s the loop. Once you trust it on small things, hand it bigger ones.

We’ll be shipping fast. Join the Discord and tell us what’s broken.