Build, refine, save, and share small web tools with AI.

1. Describe the tool. Start on the homepage with a direct prompt like “Build a mortgage calculator for first-time buyers with taxes, insurance, and a downloadable summary.”
2. Wait for the create page. Tool AI signs you in anonymously if needed, sends the prompt, and streams a generated project into the preview.
3. Refine with follow-ups. Ask for scoped changes such as “Keep the layout, add validation and a mobile-friendly summary.”
4. Save or download. Anonymous users get local autosave. Signed-in users can save to cloud projects and optionally publish to the gallery.

Tool AI supports two web endpoints. The AI can request these on the server before generation, and generated tools can also call them at runtime when a tool needs search or documentation lookup.

/api/search-web?query=MDN Fetch API documentation&count=3

/api/read-web?url=https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API&maxChars=8000

Generated browser tools can request API keys, use the visitor's Tool AI account, and store shared project data. Every capability is explicit: the tool references a client script or global, and the trusted Tool AI host handles credentials, dialogs, and permissions.

async function initialize() {
  const MANUAL_API_KEY = "";
  const key = MANUAL_API_KEY || await window.toolaiKeys.get("google_maps");
  // Use the key only with the intended provider.
}
initialize().catch(() => {
  // Show setup instructions without logging credentials.
});

<script src="/api/tool-auth/client.js"></script>

const user = await toolaiAuth.ready();        // null for guests
await toolaiAuth.signIn();                    // host confirms, then redirects
await toolaiAuth.storage.set("notes", [...]); // private, per tool

<script src="/api/cloud/client.js"></script>

await cloud.ready();
const views = await cloud.increment("my-project:views", 1);
await cloud.set("my-project:items", [{ title: "Shared item" }]);

Strong prompts describe the user, output, required controls, data states, and constraints. Follow-ups should say what must stay unchanged.

Build a [tool type] for [target user].
Must include: [feature 1], [feature 2], [feature 3].
Data states: empty, editing, validation error, success.
Style: [layout], [tone], [visual constraints].
Behavior: [exact interactions and edge cases].
Keep unchanged: [things the AI must preserve].

Pick a Theme in the create toolbar (or beside the prompt on the home page) to apply a preset design system to the generated tool. The preset supplies its own classes, variables, and components, so the tool's style.css holds only the additional or overriding CSS the request needs.

• Default — no preset; the tool generates its complete stylesheet.
• Blueprint, Glasswave, Midnight, Modern, Paper, Playful, RetroOS, Soft UI, Studio, and Terminal presets.
• In the prompt, describe styling concretely — for example “warm editorial paper with terracotta accents” or “brutalist buttons with hard offset shadows.”
• Switching back to Default removes the preset CSS instead of leaving it in the project.

Choose a model from the selector on the create page, or manage your personal list in the Model Shop. Model preferences are stored in this browser.

In the editor, choose Export to GitHub beside Download ZIP. Signed-in users create a new repository in their personal GitHub account and push the current source files, including unsaved edits and nested folders. This exports source only; Tool AI cloud storage and hosted services are not migrated or deployed.

• Choose private (the default) or public repository visibility.
• Provide a GitHub classic personal access token with repo scope, or public_repo for public repositories only.
• The token is used for that export request only and is never saved or logged. No server-side GitHub configuration is required.
• Up to 500 text files and 900 KB of serialized source.
• Exports never overwrite an existing repository. If an upload fails after creation, open the repository to inspect it and retry with a new name.
• Export does not establish ongoing sync with GitHub.

Settings → Connections lets you authorize outside AI clients to reach your existing Tool AI account and projects. Tools and projects are the same resources you see in Tool AI.

How spark cost is estimated

The server estimates one input token per four characters of prompt, project source, and conversation history. Cost is the larger of 500 or the estimated input tokens, multiplied by the model cost, with a minimum charge of 500 sparks. Model multipliers range from 0.6x to 3.5x; unknown models use 1.2x. A short request costs 500–1,750 sparks, allowing up to 80 minimum-cost requests per day and 40 per session for signed-in users. Larger projects cost more. Signed-in users share 200,000 sparks per UTC day and 600,000 per UTC week. Guests retain 4,000 daily, 12,000 weekly, and 2,000 per session, with a separate shared pool of 10,000 daily and 30,000 weekly. DeepSeek also has separate shared spending limits. These are conservative app allowances, not a guarantee of provider availability. Failed generations still consume their reservation because providers may have used quota.

Other limits

• Prompt text area: 4,000 characters.
• Saved message history: up to 100 messages.
• Generated project: up to 80 files.
• Single generated file: up to 300,000 bytes.
• Total generated project payload: up to 750,000 bytes.
• Temporary streamed preview lifetime: about 10 minutes.
• Web page reader: up to 12,000 returned characters.
• Web fetch body cap: 1 MB.
• Web reader timeout: 10 seconds.
• AI-requested web tool calls: up to 4 per generation.
• NVIDIA server-side guard: up to 40 requests per minute by default.
• API key entry: up to 4,096 characters; service names use lowercase letters, digits, and underscores (up to 64 characters).
• Private tool storage: 64 KiB per signed-in user per tool; keys up to 200 characters.
• Shared cloud values: keys up to 100 characters, and up to 100 keys in one state read.
• GitHub export: up to 500 text files and 900 KB of source.

• Generated HTML passes through a light markup filter before preview and save. It intentionally preserves executable code, inline handlers, and modern CSS, so treat generated tools like any untrusted web page.
• Executable URL schemes such as javascript: are filtered from links, scripts, and forms.
• Project file paths must be safe relative paths, and file and project size limits are enforced.
• Tool AI cannot be embedded by other sites because frame ancestors are restricted to itself.
• The web reader blocks localhost, private IP ranges, and private redirect targets.
• Cloud project saving requires a verified, non-anonymous Firebase user.
• Private tool storage requires a signed-in user and is scoped to the verified user and tool.
• Public gallery publishing requires ownership of the saved project.

Can I edit an existing project?

Yes. Open a saved project from Projects and continue with follow-up prompts. The current files and prior conversation are included as context.

Can Tool AI read documentation?

Yes. It can search for public pages and read public documentation text before generation. It cannot read private sites, localhost, or private-network URLs.

Why did the AI redesign something I wanted to keep?

Follow-ups should explicitly say what must remain unchanged. For example: “Keep the current layout and calculations. Only add error messages.”

Can a generated tool use my Tool AI account?

Yes. With the auth client, a tool can show sign-in and sign-out prompts, read the signed-in profile, and store private data per user and per tool. Guests appear as null, and the tool never receives tokens, passwords, or Firebase configuration. Account features work in the Tool AI editor and viewer, not in standalone exports.

Can generated tools share data between visitors?

Yes, through the shared cloud client. Values are public to anyone who knows the key, so prefix keys by project and never store private records or secrets there. Use account private storage for per-user data.

How do I connect ChatGPT or another AI client?

Use GPT Connect or Connections in Settings. You authorize the client with a normal browser sign-in, review the requested permissions, and can revoke access at any time.