# Toone

> A native macOS workspace for building, running and supervising repeatable AI agent routines with checks, memory and pause-and-resume control.

Toone is a native macOS workspace for projects that use AI agents. Users describe a workflow in plain words, and Toone drafts a graph of steps that specialised agents run on the models and accounts the user chooses. It works with the user's own OpenAI or Anthropic subscription.

## What It Is

Toone keeps specialised agents, the repeatable routines they run, and the tools and knowledge they share in one place. According to the site, it is aimed at operators, researchers, marketers, founders and creators with multi-step work that recurs weekly or monthly. Users bring their own Claude Code or Codex and turn the work into a routine they can watch, pause, edit and review.

## How a Routine Runs

A routine is described in natural language, including its schedule, models and accounts. Agents run in parallel, each on a chosen model, and each step shows live progress on the user's own accounts. Contracts check every output, so a step that falls short pauses on its own. The user can then edit it and resume where it stopped. Failure routes can retry, switch model, or ask the user, and routines can pause for approval before sensitive steps such as publishing.

## How It Differs From a Scheduled Prompt

Toone's site contrasts routines with a prompt on a timer. A routine is a visible graph of steps with memory between runs, so it remembers what it already did, such as leads already found. It also supports a different model per step.

## Features and Setup

The site lists a Spotlight-style summoner for agents, resizable workspace windows, browser sessions for existing tools and accounts, voice input, and meeting recording with speaker labels. Live collaboration and a human-control feature are marked beta. An Explore library offers reviewed routines to add and adapt. A guided quickstart takes about 15 minutes from install to a first routine run.

## Features
- Describe workflows in plain language and get a draft step graph
- Parallel agents with a model chosen per step
- Contracts that check outputs and pause failing steps
- Pause, edit and resume in place
- Failure routes: retry, switch model, or ask the user
- Memory between runs
- Scheduled routines
- Approval prompts before sensitive steps
- Spotlight-style agent summoning
- Browser sessions using your own accounts
- Voice input and meeting capture with speaker labels
- Explore library of reviewed routines
- Live collaboration (beta)

## Integrations
Claude, OpenAI, Claude Code, Codex, Figma, LinkedIn, Threads, Facebook, Instagram, Higgsfield

## Platforms
WINDOWS, MACOS, LINUX, WEB

## Pricing
Subscription-based

## Links
- Website: https://trytoone.com
- Documentation: https://trytoone.com/en/how-to
- EveryDev.ai: https://www.everydev.ai/tools/toone
