> For the complete documentation index, see [llms.txt](https://docs.apolo.us/index/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.apolo.us/index/apolo-mcp/guides/workflows.md).

# Workloads and Flow

Apolo MCP supports bounded research jobs and Apolo Flow live jobs and batch bakes. Use the native tools for structured lifecycle operations; keep interactive or high-volume terminal workflows in the user's local CLI.

## Research jobs

1. Resolve and display the exact context, then use `list_presets` to select compute.
2. Review the image, preset, command, mounts, secret reference names, lifecycle limits, target context, and expected output location before starting the job.
3. Run with an explicit lifespan and scheduling timeout. Record the returned job ID.
4. Monitor with bounded wait, logs, and telemetry calls. Preserve terminal state, truncation markers, and exit information.
5. Verify the expected output and delete only an exact resource created by the current workflow and recorded in its ownership ledger.

Credentials belong in named secret references, never commands, ordinary environment values, evidence, or logs. Move large or binary data through mounted storage or a user-reviewed CLI transfer rather than through model context.

## Apolo Flow

For a Flow project, first inspect its configuration and validate the requested project, action, parameters, executor, images, volumes, and outputs. Project and configuration paths must remain under the allowed workspace.

Use the typed live-job operations to list, run, inspect, read bounded logs, wait, and kill. Start batch bakes through Flow orchestration, then inspect attempts and tasks with the typed bake operations. On failure, preserve the failed state and logs, identify the specific source or parameter correction, and use the supported restart operation only after that correction is reviewed.

Interactive attach, port forwarding, Flow project scaffolding, bulk upload/download, and local cache or project maintenance remain local CLI activities. MCP does not wrap those streams as ordinary tool results.

See the [research-job and Flow skills](/index/apolo-mcp/capabilities/skills.md) and the exact [tool reference](/index/apolo-mcp/capabilities/tools.md).