Documentation
This directory is the Marigold documentation. It is read here on GitHub, and published at marigold.run/docs. A file with front matter is published; a file without it is not.
Start
- Getting started – install the CLI, run a first application, call the OpenAI-compatible endpoints.
Applications
- Applications – what an application is, and the three layers it runs in.
- Building –
marigold.toml,models.yaml, the instance contract, packaging. - Porting – bringing an existing AI application to Marigold.
- Publishing – registries, and publishing one with GitHub Pages or any static host.
- Installing – adding registries, installing by name, dependencies.
- Finding – registries, catalogues, and the index for tools and agents.
- Examples – the examples registry.
- Showcase – the showcase registry.
Reference
- Configuration –
marigold-host.toml: every host setting, output storage, resource limits, authentication. - Deployment – from a laptop to your own cloud account: managed Postgres, GPU worker hosts, S3 and Azure storage.
- Model types – every model type, its input and output.
- API – submitting and polling, binary outputs, the OpenAI-compatible endpoints.
Design
- Architecture – how the running system is divided, and what each part may touch.
- Principles – the reasoning behind recurring design decisions.