Experimental — This project is under active development and not yet production-ready. APIs, data formats, and architecture may change without notice.

An AI-powered assistant for building and editing JSON Schema + UI Schema form definitions through natural conversation. Describe what you want your form to look like — in German or English — and the agent issues precise, structured edits to the live schema document pair.

• Conversational form authoring — users describe fields, validation rules, and layout in plain language; the agent translates intent into exact schema mutations
• Human-in-the-loop clarification — when intent is ambiguous (e.g. "Dropdown" could mean oneOf, autocomplete, or an API-backed lookup), the agent pauses and asks rather than guessing
• Minimal, auditable edits — every change is a targeted patch against the current schema version; inverse patches are stored so undo is always possible
• Multilingual — responds in the user's language (German or English), including domain-specific vocabulary like Pflichtfeld (required field) and Adresseingabe (address sub-schema)
• Embeddable UI — a draggable floating chat component (AgentFAB) that can be dropped into any React app alongside an existing form renderer

apps/
  vercel-schema-agent-server/   # Hono API — streaming AI agent, Redis session store
  storybook/                    # Component development and integration demos

packages/
  agent-chat-components/        # Headless-ish MUI v7 chat UI (no backend dependency)
  agent-chat-flow/              # useChat wiring, schema sync, clarification handling

The server exposes three HTTP endpoints:

The agent runs on Anthropic Claude (default) or a local Ollama model, configured via environment variables.

• Nix with flakes enabled, or Bun ≥ 1.1 and Redis installed manually
• An Anthropic API key (or a running Ollama instance)

nix develop

This provides bun and redis, and registers the run-dev-redis helper command.

bun install

cp .env.example .env
# Edit .env and set ANTHROPIC_API_KEY (or set PROVIDER=ollama and OLLAMA_BASE_URL)

run-dev-redis   # Nix dev shell
# or
redis-server    # if installed manually

bun run dev:server
# Server listens on http://localhost:3001

bun run dev:storybook
# Opens on http://localhost:6006

# Create a session
SESSION=$(curl -s -X POST http://localhost:3001/api/session | jq -r '.sessionId')

# Send a message (streams tool-call events + text)
curl -N -X POST http://localhost:3001/api/chat \
  -H "Content-Type: application/json" \
  -d "{\"sessionId\":\"$SESSION\",\"message\":\"Füge ein Pflichtfeld für den Vornamen hinzu\"}"

# Inspect the schema after the edit
curl http://localhost:3001/api/schema/$SESSION

MIT