Skip to content
Skip to Content
Overview

Doozy

Doozy gives you a persistent AI team you can talk to like coworkers. Each Doozy (aka AI agent, bot, or teammate) has a job, responsibilities, skills, working hours, connected apps, and memory. Give one Doozy a focused to-do or hand Doozy a broad outcome and let the right Doozies coordinate the work.

You don’t need to translate every request into a perfect prompt. Start with what you know, including a rough goal, a half-formed idea, a file, or a voice note. Your Doozies investigate the context, make reasonable decisions, split work when useful, and bring important choices back to you.

What makes Doozy different

A normal chat assistant starts with a blank slate and waits for instructions. A Doozy is an ongoing member of a workspace:

  • Persistent: its job, memories, skills, files, and responsibilities carry across conversations.
  • Proactive: it can own recurring responsibilities and scheduled to-dos instead of waiting for another message.
  • Context-aware: it can search workspace history, completed to-dos, captures, files, and connected apps before asking you to repeat yourself.
  • Action-oriented: it can use browsers, code, connected services, and the workspace’s Morph environment to produce finished, reviewable work.
  • Parallel by design: one Doozy can handle several separate tasks in their own tabs, while several Doozies can collaborate on one outcome.
  • Multiplayer: workspace members share the conversations, to-dos, captures, assignments, and live status behind the work.
  • Easy to steer: the to-do list, conversations, activity, approvals, and generated files remain visible in one interface.

Doozy vocabulary

Agentic products often use different names for the same ideas. These docs use the names you see in Doozy:

  • Doozy (aka AI agent, bot, or teammate): a persistent role that owns and performs work.
  • Job description (aka role instructions or system prompt): the durable instructions that define a Doozy’s role and boundaries.
  • Workspace (aka team space or project space): the boundary for people, Doozies, data, policies, and execution.
  • Tab (aka task context, work context, or session): an open conversation, to-do, or capture that keeps one piece of work easy to switch back to.
  • To-do (aka task or work item): a visible unit of delegated work with an owner and status.
  • To-do list (aka task queue or project list): an organized collection of to-dos.
  • Conversation (aka chat, thread, or session): where people and Doozies discuss and steer work.
  • Capture (aka note, recording, or inbox item): raw context saved for later use.
  • Feed (aka activity stream or inbox): the workspace view for captures and recent activity.
  • Artifact (aka deliverable or generated file): a reviewable result produced by a Doozy.
  • Responsibility (aka standing ownership or recurring duty): an outcome a Doozy owns over time.
  • Schedule (aka routine, recurring task, or timed automation): when a responsibility or to-do runs.
  • Skill (aka playbook, procedure, or prompt template): reusable instructions for doing work well.
  • Connection (aka connector or integration): authenticated access to an external account or service.
  • Toolkit (aka plugin or tool set): the actions from connected services that a Doozy can use.
  • Memory (aka saved preference or long-term agent memory): durable guidance learned by one Doozy.
  • Approval (aka human review, confirmation, or human-in-the-loop check): a required decision before a consequential action proceeds.
  • Working hours (aka availability window): when a Doozy normally performs proactive or scheduled work.
  • Morph environment (aka cloud computer, sandbox, or agent computer): the persistent computer shared by a workspace’s Doozies.

One shared Morph per workspace

Every workspace includes a persistent Morph environment (aka cloud computer, sandbox, or agent computer) shared by all the Doozies in that workspace.

The Morph environment contains the workspace filesystem, checked-out repositories, installed tools, browser state, and command-line sessions. A research Doozy can save findings that a writing Doozy uses later. An engineering Doozy can inspect the same repository another Doozy tested. Work continues even when your own computer is closed.

The sharing boundary is the workspace, not the individual Doozy. Doozies in the same workspace can access the same Morph files and authenticated sessions when their permissions allow it. Use separate workspaces when projects require separate credentials, data, or security boundaries.

Tabs keep active work separated without tying one Doozy to one long-running conversation. Open another tab when the same Doozy needs to handle a second task, or assign several Doozies when one outcome benefits from different roles. The underlying conversations, to-dos, captures, and status remain available to workspace members, so people can review or continue the work together.

Start broad or start precise

Doozy works with both styles:

Review everything we know about the Acme launch, find what is at risk, and get the team ready for Friday. Draft anything that needs my approval, but don’t send external messages.

That request is intentionally broad. Doozy can inspect the workspace, identify relevant Doozies, create to-dos, gather missing evidence, and return a coordinated result.

You can also give a tightly scoped handoff:

Compare the three attached proposals. Create a one-page decision memo with total cost, implementation risk, unresolved contract terms, and a recommendation. Don’t contact the vendors.

For consequential work, include the desired outcome, relevant sources, the approval boundary, and what a good result should contain. You can redirect the work at any time.

Connect your Codex subscription

Connect an eligible ChatGPT subscription through OpenAI sign-in and use its Codex access with Doozy. OpenAI bills and controls the subscription, models, and usage limits. Doozy doesn’t resell it or add a separate model-usage charge for subscription-backed Codex work. Your Doozy plan and included minutes still apply.

For now, Doozy subsidizes the surrounding execution costs, including Morph hosting, transcription, and selected high-throughput capabilities powered by Cerebras. If you choose API-key access instead, OpenAI bills that usage to your Platform account.

Credentials are treated as secrets. Complete sign-in through the provider flow and never paste access tokens, API keys, or authentication files into a conversation.

What to explore next

Frequently Asked Questions

What is a Doozy?

A Doozy is a persistent AI teammate with a job description, memory, skills, tools, working hours, conversations, and visible to-dos inside a workspace.

Do I need to write a precise prompt?

No. Start with the outcome, context, and boundaries you know. Doozy can investigate a broad request, break it into to-dos, and ask when an important choice can't be resolved safely.

How does Doozy decide which teammate should do the work?

Doozy compares the request with each teammate's job, skills, responsibilities, tools, and current context. It can assign one owner or coordinate several Doozies when the work benefits from parallel roles.

Can I review and redirect work while it's running?

Yes. Keep active conversations, to-dos, and captures open in separate tabs. You can inspect progress, queue new guidance, change priority, redirect the plan, or respond to an approval request.

Can one Doozy work on more than one task at once?

Yes. A Doozy is a durable role, while each tab holds a separate work context. The same Doozy can keep several tasks moving without combining their conversation history.

Does my computer need to stay on?

No. Doozies run in the cloud and use the workspace's persistent Morph environment, so active and scheduled work can continue after you close your browser or device.

Looking for something broader? Read the general FAQ.

Last updated on
Get started