You have to leave in ten minutes. Your coding agent has found the bug, changed a few files, and started working through the failing tests. There is useful work left to do, but your laptop is about to go into a bag.
With Ondago, you can move a supported session to a cloud environment and keep working from your phone. The conversation, captured project files, and compatible setup travel together. When you return, you can bring the result back to your computer and review it.
The important moment is before you close the lid: confirm that the cloud agent can actually continue the task.
Check where the agent is running#
Opening a session on your phone does not move the agent off your laptop. These three states have different consequences when the computer goes offline:
| State | What you can do | Can the laptop go offline? |
|---|---|---|
| Local session, open on your phone | Read and control the agent running on your computer. | The computer must stay awake, online, and hosting the session for work to continue. |
| Saved snapshot | Read the captured session or use it to resume work later. | Yes, after the save completes. Saving alone does not start an agent. |
| Cloud continuation | Work with the agent running in the cloud. | Yes, after confirming the transfer and checking that its tools no longer depend on this computer. |
This walkthrough uses an explicit cloud transfer while the source computer is still online. Start it before packing up.
1. Give the task a clear stopping point#
Pick a session in Ondago Desktop and check the project it belongs to. In our example, the task is to fix a checkout validation bug and finish the tests. Give the agent an instruction that will still make sense when you are away:
Finish the checkout validation fix. Run the relevant tests, summarize what changed and any failures, then wait for my review. Do not deploy.Review cloud access for the selected agent and workspace. The destination needs working provider credentials or a connected account, access to the repository when required, and available cloud allowance. A local agent sign-in does not always provide cloud access. The supported agents guide explains compatibility, and the Codex cloud guide covers connecting a ChatGPT subscription.
Check dependencies on your laptop, too. A local database, a development server, or an MCP connection set to Use this computer still needs that computer. Arrange the access the task needs before leaving. See where MCP connections run.
2. Transfer the session and verify the destination#
- Select the session in Desktop and choose Continue in cloud.
- Review the session and project being transferred. Resolve any missing credentials, repository connection, or compatibility choice.
- Confirm the transfer and follow its progress. Keep the source computer online until Ondago shows the cloud session as ready or running.
- Open the cloud conversation. Check the latest response, and send the next instruction if the agent is waiting for you.
For the checkout task, look for evidence that the agent can read the project and continue the tests. A connected conversation alone does not establish that the provider accepts its credentials or that the selected model is available.
Once the cloud task is working and no required tool depends on your laptop, you can close it. The cloud continuation guide has the full flow.
Prefer the terminal? Choose the command for the session#
For a session that is no longer running locally, find its ID and lift it with cloud continuation:
cd ~/projects/example-app
ondago detect -plain
ondago lift -session SESSION_ID -watchReplace the example path and SESSION_ID with your project and session. To save a snapshot without starting a cloud run, add -no-cloud.
For a supported agent actively working in another terminal, use a handoff. This example uses Claude Code; check the CLI cloud guide and your installed command help for your agent:
ondago init-hooks
ondago handoff -cwd ~/projects/example-app -session SESSION_ID -tool claude-code -note "Finish the checkout tests, summarize the results, then wait for review. Do not deploy."Follow the printed transfer result and verify the cloud conversation before closing the source computer, just as you would in Desktop.
3. Follow the task from your phone#
Open app.ondago.ai, sign in with the same account, and select the same workspace. Complete device registration and workspace access if asked. It helps to connect your phone while your signed-in computer is still online.
Open the current cloud continuation in Sessions. Read the agent’s progress, answer its questions, and give the next instruction when needed. If the checkout tests reveal another failure, you can ask the agent to investigate it or stop and explain what remains.
Continue in that cloud conversation so you are working with its latest files and messages. Starting again from an older local copy creates a different continuation. Cloud runtime and provider usage limits still apply; check your workspace’s allowances before leaving a long task running.
4. Bring the work back and review it#
Back at your computer, open the cloud session in Desktop and choose Continue locally. Review the destination folder, then choose Pull. Ondago saves the latest cloud checkpoint and brings it to this computer.
For a Git repository, commit or stash local edits before applying returned work to your checkout. Review any conflicts or preserved copies reported during the return. Then inspect the changes and run the relevant checks locally before continuing with your installed agent.
From the terminal, inspect the run and collect its result bundle:
ondago runs
ondago list
ondago pull RESULT_BUNDLE_ID -dir ~/projects/example-app -branch -no-openUse the result bundle ID for the cloud work you want to collect. The branch option lets you review the returned changes separately; -no-open leaves the agent closed until you are ready. Follow the printed review and resume instructions. See bringing cloud work back for the full return flow.
What moving to the cloud means for privacy#
Ondago encrypts captured session content before uploading it. For a cloud continuation, the destination runtime receives an encrypted copy it can decrypt to do the work. The storage service does not receive your plaintext conversation, but the runtime executing the agent must be able to read the supplied project and use its credentials.
That runtime, its infrastructure, and the selected agent and model provider are part of the trust boundary for the run. Review what you are supplying and use credentials appropriate to the task. The security guide explains the distinction between encrypted storage and cloud execution.
Before you close the lid#
- The transfer completed and you opened the cloud conversation.
- The agent can respond and has a clear next task.
- Required tools and connections can work with your laptop offline.
- Your phone can open the same workspace and continuation.
Try the round trip with a small task: a failing test, a documentation edit, or a contained bug fix. Move it to the cloud, check it from your phone, then pull the result home.