Local to Cloud

This guide upgrades a local Codex workflow to the cloud while preserving artefacts and tooling. Start with a repository you already run through node ./bin/start-session.mjs or computer-agents locally.

Step 1: Capture the baseline

Run a local session and inspect docs/sessions/<id>/ to understand the artefacts your workflow relies on.
Note MCP servers in use and any environment variables the agent needs.

Step 2: Prepare the workspace

Ensure the repository is clean and pushed to git.
If you rely on secrets, move them into .env or configuration files that can be uploaded via the Files API.

Step 3: Provision a container


const cloud = new TestbaseCloud({ apiKey: process.env.TESTBASE_API_KEY });
 
const agent = await cloud.createCloudAgent({
  name: 'Worker',
  agentType: 'worker',
  workspace: './repo',
  model: 'gpt-4o-mini',
  reasoningEffort: 'medium',
  persistContainer: false
});

If you need MCP servers, add them now with cloud.addMcpServer.

Step 4: Seed configuration

Upload secrets, fixtures, or dependency manifests:


await cloud.uploadFile(agent.containerId, '/workspace/.env', process.env.LOCAL_ENV!);
await cloud.uploadFile(agent.containerId, '/workspace/scripts/bootstrap.sh', bootstrapScript);

Step 5: Execute the same task


const result = await run(agent, 'Add feature X exactly as the local workflow does');
console.log(result.finalOutput);

Compare artefacts:

cloud.listFiles to view outputs
cloud.downloadFile to retrieve logs or test reports

Step 6: Validate parity

Diff local and cloud outputs.
Ensure reviewer feedback or plan artefacts align.
Check cloud.getLogs for runtime differences (missing packages, slower commands).

Step 7: Automate cleanup

Decide whether to keep the container alive. For CI jobs, delete it:


await cloud.deleteContainer(agent.containerId);

Step 8: Update orchestrations

Point orchestrators or schedulers at the new cloud-backed agent.
Store the container ID in a secrets manager or recreate the container per run if you prefer immutability.

Following this sequence minimizes surprises: you graduate to the cloud with confidence that behaviour matches your trusted local workflow.