Bring your own brain.
Every major model API, swappable per task. Pick the agent that fits the job.
Anthropic's flagship for long-context reasoning and tool use — the default brain for agent loops.
OpenAI's multimodal workhorse — broad ecosystem, vision, and fast function calling.
Google's million-token model — ideal when you need to drop in whole repos or hours of transcripts.
Meta's open weights — run it locally for privacy or fine-tune it for niche domains.
Lean European models with strong tool use — cheap, fast, and Apache-licensed.
xAI's chatty model with real-time access to X — useful for news-aware agents and sharp tone.
Not a model — an inference provider. Runs Llama and Mixtral at hundreds of tokens per second.
Run any open model locally on your machine — fully offline, zero token costs, full privacy.
Open-source agent harness built around tool authoring and durable memory.
Nous Research's fine-tunes — agentic behavior tuned for structured output and persona.
Frequently asked
The stuff people ask before they sign up.
Froots is a local-first desktop app where you run AI agents — Claude Code, OpenAI Codex, Cursor, or your own harness — and watch every move they make. Tool calls, file diffs, browser sessions, memory updates, and reasoning all stream into one timeline you can pause, fork, and replay. You bring the model keys; your data stays on your machine.
Claude Code, OpenAI Codex, Cursor agents, and your own homegrown harness. We speak to each through a thin adapter so the interface and controls stay identical regardless of who's driving.
Shell, browsers, MCP servers, your private APIs, and anything you can wrap in a function. Wire a tool into a run with a few lines of config — agents call it, you see exactly what they did.
Yes. Every run is fully interruptible. Pause anything risky before it touches prod, review the diff, and either approve, edit, or kill the run. You stay in the loop on the steps that matter.
Your runs, memory, and knowledge base stay on your machine by default. Bring your own model keys; we never see your prompts or outputs unless you opt in to cloud features.
Chat UIs show the conversation. Froots shows the work — tool calls, file diffs, browser sessions, memory updates, and reasoning all streamed into one timeline you can replay and fork.
No. The starter routines, widgets, and skills work out of the box. If you're a developer, you can write your own skills and harness adapters — but most teams ship without writing code.
Free forever when you bring your own model API keys (Claude, GPT, Gemini, Ollama, local). $12/mo if you want to sync your vault and memory across devices. Beta pricing locked in for life. See /pricing for full details.