Workspace notes / Project handoff
Keep the work. Resume in a new chat.
A persistent AI agent workspace gives a project somewhere to live outside the conversation. Save the brief, decisions and unfinished work to files, then ask the next chat to read them before continuing.
Join the computer waitlistThe computer workspace is on a waitlist. Saved files provide context to read, not automatic model memory, a backup guarantee or an always-running agent.
Published agent plans are separate. See agent plans.
- First chatWrite the handoff
Save decisions and open questions to named files.
- Between sessionsFiles stay put
The workspace holds the saved project state.
- Next chatRead, then resume
Load the handoff and confirm the next action.
The practical part
Give the next chat a starting point.
Name the project files
Keep a brief for the goal, a decisions file for agreed facts and a next-steps file for unfinished work. Ask the agent to distinguish your decisions from its suggestions. File names make the handoff explicit.
Close with a checkpoint
Before ending a chat, ask it to update those files and show what it saved. Review the checkpoint for missing requirements and unsupported claims. Work left only in the conversation is not a reliable file handoff.
Start by reading
In the next chat, name the files and ask for a short recap before doing more work. If a file is missing, stop and rebuild the context with the user. Persistence does not make the model read every file or remember every conversation.
Worked example
Hand off a community workshop plan.
For this example, brief.md requires an accessible entrance, the supplied venue notes leave capacity unconfirmed and no venue has been selected. Use your own notes in the connected workspace. This is an illustrative handoff, not a customer result or a tested workshop-planning run.
Ask for this
Read workshop/brief.md and the venue notes I provided. Save agreed requirements to workshop/decisions.md and unanswered questions to workshop/next-steps.md. Label suggestions as unapproved. Do not contact venues or make bookings.
Illustrative output
workshop/decisions.md - Requirement: an accessible entrance. - No venue selected yet. workshop/next-steps.md - Confirm capacity with the organizer. - Suggestion, unapproved: compare two venues after capacity is known.
In the next chat
Read workshop/brief.md, workshop/decisions.md and workshop/next-steps.md. Recap the agreed requirements and open questions. Ask me for the missing capacity before proposing a venue. Do not assume you remember the previous chat.
Check the result. Check the saved files before closing the first chat. In the next one, verify that the recap includes the capacity question and has not turned the unapproved suggestion into a decision. Export important work separately.
Make the call.
A good fit when
You return to the same project across chats and want a file-based handoff you can inspect, revise and export.
Consider another setup when
You need guaranteed long-term retention, automatic recall of every chat or a backup and disaster-recovery service.
Before you choose
Good questions.
Is a persistent workspace the same as model memory?
No. Files hold saved information. A new chat must read the relevant files and can still misunderstand them. Ask for a recap and check it before continuing the project.
Does the agent keep working while the computer sleeps?
No background work is promised here. The recorded demo shows a computer sleeping between requests and waking for a later call. This page describes resuming work when you ask.
Are the files guaranteed to stay forever?
No permanent retention or backup guarantee is made here. Keep independent exports of important work and confirm the applicable access and data terms before relying on a workspace.
How can I get this workspace?
Join the computer waitlist on the homepage. The published agent plans are separate. The recorded transcript demonstrates a file surviving between two chats; it does not establish general availability or a retention guarantee.