Core concepts

The architecture is the source of truth

In Skeema, the system architecture is the canonical model of your system. Every other artifact — the ER schema, the sequence flow, the generated code — is derived from it. This is the opposite of most tools, where each diagram is a disconnected drawing that drifts out of date the moment you change something. Because the architecture holds the real services and their relationships, Skeema can reason about it: simulate it, cost it, and project it into other views.

Derivations

A derivation is a new board generated from a node (or set of nodes) in the architecture:

• •A database node derives an ER schema and schema code (Prisma, SQL, Mongoose).
• •A service node derives API routes in the language of its tech stack.
• •A flow across services derives a sequence diagram.

Each derived board records where it came from — its parent board and source node. A small mint dot on a node tells you it already has derivations.

Lineage

Lineage is the link between a source node and the boards derived from it. Skeema tracks lineage so your views never silently drift. If you delete the node an ER schema was derived from, the schema isn’t broken — it’s flagged with a banner so you can keep it, regenerate it from a new source, or remove it. This is the foundation of Skeema’s cross-artifact consistency.

Projects

A project bundles a root architecture board with all its derived views and flows. Open the project page to see the root, the flows, and the derived boards in one place — and generate a single Markdown document describing the whole system.

How AI generation works

Generation is a pipeline, not a black box:

Type detection

A keyword-weighted matcher decides whether you’re asking for architecture, ER, sequence, agent, or pipeline — you can override it with the type chip.

Prompt enrichment

Scale hints (“500k users”) are expanded into explicit infrastructure requirements before the model sees them.

Domain templates

Recognized domains (e-commerce, SaaS, streaming, ridesharing) inject a required-components list to bias the model toward realistic infrastructure.

Sanitization

Every generator normalizes the model’s output — injecting missing primary keys, removing invalid edges, de-duplicating hallucinated branches — before it reaches the canvas.

Design language

Skeema uses color to encode meaning: mint marks connected-diagram features (derivations, lineage), violet marks AI features, and emerald marks manual edits. Node type chips (CLIENT, GATEWAY, SERVICE, DATABASE, QUEUE, CACHE) make the role of each box readable at a glance.