AI-generated websites—create projects, let AI build pages, and host them on the platform. The client.web SDK subclient powers it all.

Web

Describe your site. AI generates it. We host it. From idea to live website in minutes.

Web projects let you create AI-generated websites on Upstreet. Define a project with a name and prompt—AI produces HTML, CSS, JavaScript, and images on demand. Pages are hosted and served by the platform. Perfect for landing pages, portfolios, documentation, and NPC-authored sites.

AI-generated website preview with modern design—hero section, cards, and dynamic content

Web Projects Overview

Concept	Description
Projects Page	Main hub at `/web`—list, create, and manage web projects
Project	A named site with a project prompt; AI uses it to generate pages
Assets	Pages and files (HTML, CSS, JS, images) produced by AI
Hosting	Platform serves content at `/api/web/[projectId]/serve/[path]`
SDK	`client.web` subclient for projects and assets

Web projects use AI to generate content on the fly. Request a path (e.g. /index.html) and if it doesn't exist, AI generates it based on the project prompt. Once generated, assets are cached and served quickly.

Web Projects Page (`/web`)

The web page at /web is your control center:

Action	Description
List Projects	See all your web projects (name, description, visibility)
Create Project	Add a new project with name, description, and optional visibility
Open Project	Enter a project to browse assets, edit settings, or preview
Settings	Configure project prompt, text model, image model, visibility

Placeholder: Web projects list with AI-generated site previews

Creating AI-Generated Websites

1. Create a project

Create a web project with a name and optional description. The project prompt (set in settings) guides AI when generating pages.

import { PUClient } from 'pu-client';

const client = new PUClient({ apiKey: process.env.PU_API_KEY! });

const project = await client.web.projects.create({
  name: 'My Portfolio',
  description: 'A personal portfolio site',
  visibility: 'public',
});

2. Set the project prompt

Update the project with a prompt that describes the site's purpose, style, and structure. AI uses this when generating pages.

await client.web.projects.update({
  id: project.id,
  project_prompt: 'A minimal portfolio for a 3D artist. Dark theme, hero with work samples, project grid, contact section.',
});

3. Request pages

Request a path like /index.html or /about.html. If the asset doesn't exist, AI generates it. If it exists, it's served from cache.

const asset = await client.web.assets.get(project.id, '/index.html');
// asset.status: 'pending' | 'generating' | 'completed' | 'failed'

4. Serve content

Fetch the rendered content for a path. Use this for iframes, SSR, or direct links.

const res = await client.web.getContent(project.id, '/index.html');
// res is a fetch Response (HTML, CSS, etc.)

Project Pages and Settings

Each project has assets—files at paths like /index.html, /styles.css, /about.html. List assets with client.web.assets.list(projectId). Get a specific asset with client.web.assets.get(projectId, path). Requesting a missing path triggers AI generation.

Project settings include project_prompt (guides generation), text_model, text_model_reasoning, image_model (optional overrides), and members (viewers, editors, owners). Update via client.web.projects.update().

Projects can be public or private. Public projects appear in client.web.projects.listPublic(). Private projects require authentication to access.

Setting	Description
`project_prompt`	High-level description of the site; AI uses it for all page generation
`text_model`	Optional override for text generation model
`text_model_reasoning`	Optional reasoning model for complex pages
`image_model`	Optional override for image generation
`visibility`	`public` or `private`
`members`	User IDs and roles (viewer, editor, owner)

How Websites Are Hosted and Served

flowchart LR
    subgraph client [Client]
        Request["Request /path"]
        SDK["client.web"]
    end

    subgraph api [API Layer]
        Assets["Assets API"]
        Serve["Serve Endpoint"]
    end

    subgraph ai [AI Pipeline]
        Gen["AI Generation"]
        Cache["Asset Cache"]
    end

    Request --> SDK
    SDK --> Assets
    Assets --> |"Asset exists?"| Cache
    Cache --> |No| Gen
    Gen --> Cache
    Cache --> Serve
    Serve --> Client

Step	Description
1. Request	Client requests `get(projectId, path)` or `getContent(projectId, path)`
2. Lookup	API checks if asset exists at path for project
3. Generate	If missing, AI generates based on project prompt and path
4. Store	Generated asset is stored and cached
5. Serve	Content returned via `/api/web/[projectId]/serve/[path]`

Serve URL format:

GET /api/web/{projectId}/serve{path}

Example: /api/web/abc123/serve/index.html returns the HTML for the project's index page.

SDK Subclient: `client.web`

The client.web subclient exposes the full web project lifecycle:

Method	Description
`client.web.projects.list()`	List all your projects
`client.web.projects.listPublic()`	List public projects
`client.web.projects.getById(id)`	Get project by ID
`client.web.projects.getByName(name)`	Get project by name
`client.web.projects.create({ name, description?, visibility? })`	Create a new project
`client.web.projects.update(params)`	Update project settings
`client.web.projects.delete({ id })`	Delete a project
`client.web.assets.list(projectId, opts?)`	List assets in a project
`client.web.assets.get(projectId, path)`	Get asset (triggers generation if missing)
`client.web.assets.delete(projectId, path)`	Delete an asset
`client.web.getContent(projectId, path)`	Fetch raw content (Response)

// Full workflow example
const client = new PUClient({ apiKey: process.env.PU_API_KEY! });

// Create and configure
const project = await client.web.projects.create({
  name: 'Doc Site',
  visibility: 'public',
});
await client.web.projects.update({
  id: project.id,
  project_prompt: 'Technical documentation site. Clean layout, code blocks, search.',
});

// Trigger generation and fetch
const asset = await client.web.assets.get(project.id, '/index.html');
if (asset?.status === 'completed') {
  const res = await client.web.getContent(project.id, '/index.html');
  const html = await res.text();
}

Asset Lifecycle

Status	Description
`pending`	Asset requested but not yet started
`generating`	AI is producing the asset
`completed`	Asset ready; URL available
`failed`	Generation failed; check `error` field

Routes Quick Reference

Route	Purpose
`/web`	Web projects list and management
`/api/web`	REST API base for web projects and assets

Web