Mosaic

Formerly Hippo. The component was renamed to Mosaic (ADR-0004): the distribution is datahelix-mosaic, the import package is mosaic, and the CLI is mosaic. Existing code and deployments keep working through a deprecation window — import hippo, the hippo command, hippo.yaml, HIPPO_* environment variables, and hippo.* entry-point groups are all still honored (with deprecation warnings). See Installation for the upgrade notes.

A runtime for LinkML schemas. Point it at a schema and you get a typed Python SDK, a REST API, a relational database with append-only PROV-O provenance, and CEL-based dynamic validation — without writing any of that infrastructure yourself. The schema defines the application; Mosaic runs it.

Start here

• Why Mosaic? — the elevator pitch and core idea.
• Introduction — what Mosaic does and how it fits together.
• Installation — install from PyPI or source.
• Quick Start — your first schema, store, and query in 10 minutes.

Build

• Schema Guide — author entities, attributes, and relationships in LinkML.
• Data Model — how schemas map to relational storage and the graph-shaped API.
• Configuration — mosaic.yaml, storage backends, and adapters.

Use

• CLI Reference — mosaic command surface.
• TUI — interactive terminal browser (mosaic tui) over the SDK or REST API.
• API Reference — MosaicClient Python SDK.
• Typed Client — schema-derived typed client.

Extend

• Reference Loaders and Writing a Reference Loader — pull static reference data into Mosaic.
• Writing a Recipe, Installing Recipes, Recipe Reference — package and share schema fragments.

Background

• Design Principles — what Mosaic is opinionated about, and why.
• Comparison — how Mosaic relates to LinkML, Datasette, and friends.

---

Mosaic is open source under the MIT license. Source at github.com/BU-Neuromics/hippo.