Tool AI • Documentation

Build websites with prompts in minutes.

Tool AI turns plain-English prompts into runnable HTML/CSS/JS tools, then lets you iterate through follow-up prompts and save your work.

Start Creating View Projects

Quick Start

  1. Open the homepage and type your first prompt.
  2. Press Enter (or click submit) to open the create page.
  3. Wait for generation. You will get:
    • A live preview of your generated tool.
    • A short explanation in the chat panel.
  4. Add follow-up prompts to refine behavior, style, or layout.
  5. Click Save Tool to store your project.

How the Site Works

1) Prompt intake

Your prompt is captured from the homepage and forwarded to the create flow.

2) AI generation

Tool AI sends your prompt (plus any existing code) to your selected provider.

3) Safety + sanitization

Generated HTML is sanitized before rendering in the preview iframe.

4) Iteration + save

Follow-up prompts modify your current build. Saved tools can be revisited from your projects page.

Prompting Guide (High Quality Results)

Best results come from specific outcomes, constraints, and user intent.

Recommended prompt structure

Build a [tool type] for [target user].
Must include: [feature 1], [feature 2], [feature 3].
Style: [color mood], [layout], [tone].
Behavior: [exact interactions and edge cases].
Accessibility: keyboard support and strong contrast.

Great starter prompts

“Build a mortgage calculator for first-time home buyers with taxes, insurance, and downloadable summary.”

“Create a meal planner where users pick dietary rules and export a weekly shopping list.”

“Make a SaaS pricing estimator with plan toggles, seat counts, and annual discount logic.”

Follow-up prompt patterns

  • “Keep everything, but improve spacing and typography.”
  • “Add form validation with clear inline error messages for each field.”
  • “Do not replace current layout. Add dark mode toggle with state memory.”
  • “Improve mobile UX below 768px and keep desktop unchanged.”

AI Providers

Gemini (default)

Fast default path with automatic server-side orchestration.

Puter.js

Uses Puter authentication and supports selecting a Puter model.

Ollama route

Uses a proxy chat endpoint with fallback behavior when needed.

Safety, Limits, and Guardrails

  • Prompt length guardrail: 4000 characters per message.
  • Conversation history trimming is supported when chats get long.
  • HTML output is sanitized before preview to reduce unsafe content.
  • Unsafe prompts can be rejected, and usage limits may apply by account state.

Error Handling & Troubleshooting

If something fails, Tool AI surfaces useful errors in the UI so you can recover quickly instead of losing progress.

Issue What it means What to do
Prompt is too long Your message is above the 4000-character limit. Shorten the prompt. Break one large request into smaller, iterative follow-ups.
Usage limit reached Request quota has been exhausted for the current period. Wait for reset, reduce prompt retries, and batch edits into one focused request.
Puter sign-in failed Provider authentication did not complete successfully. Re-select Puter in settings and complete sign-in again.
No tool to save There is no valid HTML in the preview yet. Generate once first, then click Save Tool.
Project not found / permission error The project was removed or belongs to a different account. Open your projects list and load a project owned by your current session/account.

Reliability tips

  • Keep prompts specific and scoped to one improvement at a time.
  • If a response is conversational (no code), ask directly: “Return updated HTML/CSS/JS only.”
  • Use follow-up prompts like “keep current layout unchanged” to avoid accidental redesigns.

Why Tool AI Can Feel Faster Than Base44-Style Builders

If your goal is rapid prototyping with direct control over HTML, Tool AI is optimized for short prompt-to-preview loops.

Direct code output

You get runnable HTML/CSS/JS immediately, so there is less translation between a visual builder model and code.

Fast iteration loop

Follow-up prompts can patch existing code quickly (“add X, keep Y”), which reduces rebuild time.

Provider flexibility

You can switch between provider routes (Gemini, Puter, Ollama) based on speed, style, or availability.

Built for tool generation

The prompting workflow is tuned for generating small web tools quickly, not navigating complex builder UI layers.

Note: “Faster” depends on use case. Teams generally see gains when they prefer prompt-driven coding over drag-and-drop composition.

FAQ

Why does my result not match what I imagined?

Add explicit constraints: exact sections, spacing, interaction rules, and what should stay unchanged.

Can I edit an existing project?

Yes. Open a saved project, then continue with follow-up prompts.

What if I only want explanation, not code?

Ask directly in the prompt. The assistant can return conversational guidance when code is not required.