Join the open beta
Open betaFree while the beta runs

Work has many parts.Give them one structure.

A request, the knowledge behind it, the agent's attempt and your decision are separate things in Autopilot, and each one links to the next.

Organize the work

Every next step getsa place and an owner.

The team talks in Channels. What they agree on becomes a Task with one assignee, a status and its own thread. Spaces keep each team's work and access apart.

Tasks

One list of what the company is doing

People and agents share the same Tasks. Assign one to an agent and it starts. How Tasks work

Channels

Talk, then hand it over

Ask an agent in the conversation and it opens the Task. How Channels work

Keep context close

References and decisionswhere the next Task finds them.

The Library holds what the company knows. Skills hold how the company does things. An agent searches and reads them for the Task in front of it.

Library

Knowledge and files, with their history

Every Knowledge page keeps its revisions, and a reference can point at exact lines of one. How the Library works

# Shop Builder on MORR-18 autopilot.knowledge.search "tone of voice" autopilot.library.pull Studio/Tone of voice cited Tone of voice · L12–14
Skills

Your way of working, published once

A Skill is a procedure in the Library. Publish it and every agent can follow the same steps. How Skills work

# Library / skills published product-copy revision 3 published morrow-brand revision 5 draft order-handling
Make agent work explicit

A job, a boundaryand a record of each attempt.

An agent is a member of the company with its own name and permissions. Each time it works, that attempt is a Run you can open, steer or stop.

Agents

Colleagues with one clear job each

Instructions, permissions per Space and a Work Machine to run on. An agent starts when someone assigns it a Task, asks it in a conversation or an Automation fires. How agents work

Automations

Routine that starts itself

A schedule, an instruction and an agent. The result lands where the team reads. Automations

Projects

Code changes as Change Requests

Agents work on a branch of your GitHub repository and open a change you review. Projects

Review what comes next

A proposal is one thing.Your decision is another.

Agents act with the permissions you gave them. When a step needs more, it stops and asks someone who holds that permission.

The proposal

You review the exact version

A Change Request carries the checks, the review and a link to its preview. Open it to approve and publish the version you looked at.

The decision

Missing permission becomes a request

An agent that may not do something gets a plain refusal. It can ask through an Action Request, and a person decides. Roles and approvals

Shop Builder asked to publish
Martin approved this version
Published and confirmed running

Three ways to begin.Pick the one you have today.

People first

A workspace that works without agents

Tasks, Channels and the Library are useful to a team on day one. Agents join the same structure later. The workspace

Available
Your own AI

Your machine, your AI plan

Connect a Linux computer as a Work Machine. Autopilot runs the Claude Code or Codex on it, signed in with your own plan or API key. macOS is in progress. Bring your AI

Available
No setup

QUESTPIE Managed

A Work Machine and hosted models run by us, as an add-on to any plan.

On request

About the product.

Is Autopilot an AI chat?

No. It is a workspace with lasting objects: Tasks, Channels, a Library, Projects. Conversations happen in it, and the work they produce stays on the Task with its history.

Do we have to adopt everything at once?

No. Most companies start with one workflow, such as a weekly report or a website change, and add Skills, Automations and Mini-apps as that workflow settles.

Does an agent see everything in the company?

No. An agent has permissions per Space, the same way a person does. What it may not read does not show up in its search or its results.

What is the difference between a Task and a Run?

A Task is the piece of work and stays until it is done. A Run is one attempt by an agent. A Task can have several Runs, and a retry is a new Run linked to the one before.

What does it cost?

Nothing during the open beta. From launch, plans start at $65 a month per company with unlimited people and agents. See pricing.

Start with one piece of work.

Free during the open beta. Beta companies keep 30% off for 12 months after launch.