{
  "schema": 1,
  "id": "ai-agent",
  "title": "AI Agent App",
  "description": "A chat app around an LLM agent with tool calling, retrieval over your documents and streaming answers.",
  "tags": [
    "ai",
    "llm",
    "rag"
  ],
  "thumbnail": ".w4c/preview.svg",
  "agentId": "w4c-startup",
  "defaultName": "ai-agent",
  "skills": [
    "project-bootstrap"
  ],
  "fields": [
    {
      "key": "agentName",
      "label": "Agent name",
      "type": "text",
      "required": true,
      "placeholder": "Support Copilot"
    },
    {
      "key": "stack",
      "label": "Tech stack",
      "type": "select",
      "options": [
        "Vue 3 + Quasar",
        "Nuxt 3",
        "Next.js",
        "React + Vite",
        "Angular",
        "SvelteKit",
        "SolidStart",
        "Remix",
        "Astro",
        "Laravel + Vue",
        "Django + HTMX",
        "Rails + Hotwire",
        "Phoenix (Elixir)",
        "Spring Boot + React",
        "ASP.NET Core + Blazor",
        "Go + Templ",
        "Rust + Axum",
        "Flutter",
        "React Native (Expo)",
        "SwiftUI",
        "Jetpack Compose (Kotlin)"
      ],
      "default": "Next.js",
      "required": true
    },
    {
      "key": "model",
      "label": "Default model",
      "type": "select",
      "options": [
        "deepseek-v4-flash-vision-exp",
        "deepseek-v4-flash",
        "OpenAI",
        "Anthropic",
        "BYOK"
      ],
      "default": "deepseek-v4-flash-vision-exp",
      "required": true
    },
    {
      "key": "retrieval",
      "label": "Retrieval",
      "type": "select",
      "options": [
        "pgvector",
        "Qdrant",
        "OpenSearch",
        "None (chat only)"
      ],
      "default": "pgvector"
    },
    {
      "key": "tools",
      "label": "Agent tools",
      "type": "textarea",
      "default": "search_documents, http_fetch, run_sql"
    },
    {
      "key": "features",
      "label": "Must-have features",
      "type": "textarea",
      "default": "streaming chat, document upload + RAG, tool calls, conversation history"
    },
    {
      "key": "notes",
      "label": "Additional notes",
      "type": "textarea"
    }
  ],
  "promptTemplate": "You are starting a new project from the '{{title}}' template.\n\nRepository: {{repoFullName}}\nAgent: {{agentName}}\nTech stack: {{stack}}\nModel: {{model}}\nRetrieval: {{retrieval}}\nTools: {{tools}}\nMust-have features: {{features}}\nAdditional notes: {{notes}}\n\nBefore writing any code, read README.md, SPEC.md and the .w4c/ folder: they define the scope, routes/screens, data model, flows and acceptance criteria. Then apply the project-bootstrap best practices and work top-down in layers: scaffold → streaming chat → conversation persistence → document ingestion/retrieval → tool calling → usage/limits → quality. Keep model keys server-side and stream tokens end-to-end. Keep docs/PLAN.md, the board and .w4c/project.json in sync as the project evolves. Bundled assets: {{elements}}\n\nFIRST TURN — lay the project out on its board before writing any code. Read `.w4c/boards/roadmap.json` and create one board card per layer in THIS repository (projectId = the repository full name above), prefixing every title with its unicode icon: 🧭 plan & scope, 🎨 design (mockups), 🗺️ architecture diagram, 🗄️ database & forms, ⚙️ backend & API, 💻 frontend (code), 🛡️ admin panel, 📊 control panel, 🔀 workflows, 🚀 deploy, ✅ quality & delivery. Each card body carries the goal, the deliverables, the acceptance criteria (a command or a runtime check) and the references; set the priority labels. Do not create or configure new agents — that is a follow-up, not part of bootstrap. Then report the card list and invite the user to continue on My Boards (/boards); do not start a layer before the user has reviewed the plan.",
  "elements": [
    {
      "kind": "diagram",
      "name": "Architecture",
      "path": ".w4c/diagrams/architecture.excalidraw.json"
    },
    {
      "kind": "board",
      "name": "Project roadmap",
      "path": ".w4c/boards/roadmap.json"
    },
    {
      "kind": "workflow",
      "name": "CI Build",
      "path": ".w4c/workflows/ci-build.yaml"
    }
  ]
}