Daytona

Uses a Daytona sandbox with an S3 workspace mount at options.workspaceRoot/<namespace>. The default workspaceRoot is /mnt/workspaces so Daytona matches Lambda and Kubernetes.

Configuration

{
  "name": "daytona",
  "config": {
    "provider": "daytona",
    "permissionMode": "ask",
    "options": {
      "apiKey": "...",
      "organizationId": "...",
      "apiUrl": "https://app.daytona.io/api",
      "target": "default",
      "snapshot": "fuse-s3",
      "workspaceRoot": "/mnt/workspaces",
      "mountAwsS3Buckets": true
    }
  }
}

Reference the resulting sandboxId from config.sandbox or config.workspaces[].sandbox. apiKey, organizationId, apiUrl, and target can be omitted when DAYTONA_API_KEY, DAYTONA_ORGANIZATION_ID, DAYTONA_API_URL, and DAYTONA_TARGET are set.

Use snapshot for Daytona snapshots. Use image only when creating the sandbox from a Docker/OCI image. For the current AWS-backed workspace, build the reusable mount-s3 snapshot first:

bun run daytona:s3-snapshot

sandbox.envVars passes additional environment variables into the sandbox container. All providers honor it.

network.mode maps to Daytona's networkBlockAll (allow-all → false, deny-all → true); restricted applies the CIDR allowlist only — domain allowlists are ignored with a warning.

Persistent mode (persistent: true) reserves one Daytona sandbox per workspace: idle/lifetime settings map to Daytona's autoStopInterval/autoDeleteInterval, the instance is tracked and reconnected across calls, and onCreate/onResume hooks plus detached background jobs work as described in the sandbox overview.

Requirements

Use an image or target with Node and Python installed. For persistent workspace tools, set options.mountAwsS3Buckets: true; the executor mounts the selected sandbox/<namespace>/ prefix at options.workspaceRoot/<namespace>.

What the model sees

For workspace-backed runs, the model should see a normal project directory. bash starts in the selected workspace directory:

pwd                 # current workspace directory
ls                  # files in this workspace
python3 script.py

Use relative paths in prompts and examples. options.workspaceRoot is an implementation setting for the mount location, not a path the model needs for normal file work.

Execution Notes

See Daytona runtime documentation for supported runtimes and environment setup.

TypeScript (.ts) files are not transpiled; use compiled JavaScript instead. The executor runs the tool's bash command as-is, so use python3 explicitly.

AWS S3 Workspace Mount

See Daytona's Amazon S3 mount docs for the underlying mount-s3 setup.

Set options.mountAwsS3Buckets to true to mount the current FILESYSTEM_BUCKET_NAME bucket at options.workspaceRoot/<namespace>. This is required for the Daytona provider to see the same files as the Lambda bash tool.

The snapshot must include mount-s3; use bun run daytona:s3-snapshot to build it. The script requires DAYTONA_S3_SNAPSHOT_BASE_IMAGE (base image to build from) and DAYTONA_S3_SNAPSHOT_NAME (snapshot name) to be set.

When SKILLS_BUCKET_NAME is set on the harness, the executor also mounts the account skills bucket read-only at options.skillsMountPath (default /mnt/skills); override the bucket with options.skillsBucketName.

The executor assumes the deployed sandbox-s3mount IAM role and passes the resulting short-lived, prefix-scoped credentials into the Daytona sandbox for mount-s3 — sandbox code can only reach this workspace's own key prefix and the skills bucket (read-only), never the harness runtime's broader permissions. Without the role (SANDBOX_MOUNT_ROLE_ARN unset, e.g. self-managed deployments), supply AWS_ACCESS_KEY_ID/AWS_SECRET_ACCESS_KEY through the sandbox envVars instead.

Override defaults with:

{
  "workspaceBucketName": "my-workspace-bucket",
  "awsRegion": "eu-central-1"
}

Dependencies

Use an image or image builder with packages installed before runtime.

Troubleshooting

Symptom	Cause / fix
Daytona has no available runner for snapshot '<name>'	The snapshot is non-general (pinned to one runner) or that runner is at capacity. Rebuild it as a general snapshot so any runner in the region can schedule it, drop the target region pin if it doesn't match the snapshot's region, or retry. Capacity is Daytona-side.