Skip to main content
Gremorie

Overview

Schema-driven UI for AI output. Components that consume structured LLM responses and render.

Data & visualization

Documents

Interactive

An artifact is a Gremorie primitive that takes a structured payload emitted by an LLM (validated against a schema) and renders it as UI. Charts that the model "draws". Tables that the model fills. Forms that the model designs on the fly.

Artifacts are the AI-native differentiator of Gremorie. They are the reason the registry and MCP server exist: a model that knows the artifact catalog can plan output that is structurally valid, not just textually coherent.

Anatomy

Each artifact ships with:

LayerWhat it provides
SchemaTypeScript + Zod definition for the LLM contract
RendererComponent that consumes a parsed payload and renders primitives
Streaming behaviorPartial render rules while the LLM streams
Prompt examplesHow to ask a model to emit valid output

Artifacts vs Blocks vs Components

ComponentsBlocksArtifacts
Source of stateCallerCallerLLM response (schema-validated)
Streamingn/aOptionalFirst-class -- partial renders during generation
AI awarenessNoneNoneSchema is exposed via MCP get_artifact()
ExampleButtonchat-surfacechart (LLM emits chart spec, artifact renders Recharts)

Categories

Chart

bar, line, pie, area, multi-series. Model emits a spec, artifact renders Recharts/D3.

Code

code-block, code-diff, code-runner with sandbox embed.

Document

markdown, outline, document-canvas with collaborative selections.

Data

table, list, json-tree with paging, sort, filter.

Form

dynamic form from JSON Schema, multi-step survey, slot-based custom fields.

Interactive

calculator, timeline, kanban, gallery.

One artifact per category, each demonstrating the "schema in, UI out" pattern with real previews and production-ready code.

How a model uses an artifact

The user asks: "Show me Q3 revenue by region as a bar chart."

The model calls the MCP tool get_artifact("chart") and reads the schema (categories, series, color group, axis labels).

The model emits a JSON payload that matches the schema, with the revenue figures wired into series[0].data.

The artifact validates the payload against the schema and renders a real Recharts bar chart with Gremorie tokens and chart palette.

If the payload is invalid, the artifact renders a fallback error state instead of crashing -- the schema is the contract, not the suggestion.

Status

The artifact catalog lands progressively. The chart artifact is the pilot, since the Angular edition already ships chart primitives via @gremorie/ng-data. Other categories follow as their underlying primitives ship.

For the AI primitives that compose artifacts (PromptInput, Conversation, Message, Reasoning, Tool), see Components / AI.

On this page