Agent guide

Use this page when operating from a terminal, code-generation tool, or agent without the interactive docs UI. These docs follow the current source tree and intentionally include the newest available capabilities.

Choose the route for the task

Task	Start here
Run a local evaluation	Testing locally
Generate configuration without the wizard	Getting started → Non-interactive configuration scaffold
Select storage, compute, or auth	Storage, Compute, Auth
Deploy the server	Deployment options, then Deploying
Build or call an API client	API & client and /openapi.yaml
Authenticate automation	API tokens
Diagnose a deployment	Troubleshooting and Operations

Source-of-truth order

When two summaries appear to disagree, use these sources in order:

1. CONFIG_SPEC for backend selectors and MARIMOHUB_* variables. It generates the Configuration reference.
2. /openapi.yaml for HTTP paths, schemas, authentication schemes, and response envelopes.
3. The backend pages for provisioning steps and operational constraints.
4. The platform deployment guide for topology-specific wiring.

At runtime, GET /api/v1/version reports the deployed backends and GET /api/v1/capabilities reports server limits. Prefer those responses over assuming a deployment matches a source checkout.

Non-interactive workflow

1. Select one storage, compute, and auth backend from their complete tables.
2. Copy the configuration scaffold and replace every _replace_me_ value, using the adjacent # e.g. comment when present.
3. Add backend-specific variables from the generated Configuration reference.
4. For Helm, replace <VERSION> with an existing GitHub release tag without its leading v.
5. Validate /api/health, sign in or use a PAT, create a project, start a kernel, and verify a saved notebook survives a restart.

Interactive components are conveniences only. Every required provisioning and validation step has a Markdown equivalent linked above, so raw page Markdown and llms-full.txt remain sufficient.