For the complete documentation index, see llms.txt. This page is also available as Markdown.

Apolo Platform User Context

  1. Call get_apolo_context before an operational workflow. Display the resolved authenticated username, cluster, organization, and project. Never change the user's persisted context.

  2. If any target is ambiguous, call list_clusters, list_organizations, and list_projects in hierarchy order. Pass explicit context to every later tool.

  3. Call list_presets before selecting compute. Compare CPU, memory, accelerator, scheduler/preemptibility, and credits/hour where reported; do not invent quota. If quota or price is unavailable, report it as unavailable instead of ranking on it.

  4. Resolve short storage, image, secret, and disk references with resolve_resource_uri. Do not pass a fully qualified URI outside the selected context to an MCP operation unless the user explicitly changes the target context. This is an MCP isolation boundary, not a claim that the platform cannot perform the operation. When a user asks for a supported cross-context transfer, explain the boundary and offer a CLI command for the user to review and run, for example apolo cp -r 'storage:/{source-project}/path' 'storage:/{target-project}/path' for projects in the selected cluster and organization. For other context combinations, consult the generated CLI reference, use fully qualified URIs only when the command documents them, and otherwise suggest an explicit local download/upload sequence. Never run the suggested command through MCP or imply it was validated or executed.

  5. Treat apolo as the current product name. Recognize neuro, neuro-inc, and historical platform names as retrieval aliases, not different products.

  6. Use the resource hierarchy cluster / organization / project / resource. Common resources include job, storage path, image/tag/digest, secret name, disk, bucket, app/template/revision, Flow project/live job, and bake/attempt/task.

  7. Route exact SDK signatures to SDK references, CLI flags to generated CLI pages, Flow syntax to Flow references, app journeys to GitBook, and MCP contracts to the server capability matrix. State the source and tested version when exactness matters.

  8. For an authorized administrator's read-only discovery request, use the list_admin_* and get_admin_* tools for cluster, organization, project, user, quota, balance, and maintenance metadata. These call the same Apolo SDK admin facade as apolo admin get-*; they do not grant access or perform administrative mutations. Treat user profile data as operationally sensitive and return only what the task needs.

Never request, display, or persist passed configuration, tokens, cookies, secret values, or service-account credentials. Discovery results may include identity and version metadata but never authentication material.

Last updated

Was this helpful?