Forge

Forge is the family of features that lets the model draft parts of your project for you, instead of you writing them by hand. Every Forge follows the same loop: you describe what you want, the model verifies it understands and asks about anything missing, and only when you accept does it write a file into your project. Nothing is written before you accept, and what it writes is a draft to read and refine — not a finished product.

There are two Forges today and a third on the way:

• PromptForge drafts your UI information (ui.md) from screenshots of the app under test.
• ToolForge drafts a custom tool (a .cs file) from a plain-language description.
• TestForge — coming soon — turns a narrated screencast or a written spec into a test.

All of them run on the same Forge Model (below).

The Verify loop

Each Forge page works the same way:

1. Fill in the inputs and click Verify. The model checks them and echoes back its understanding, together with the inference cost of that check.
2. If something is missing, it lists its questions under Please add. Answer them in the reply box and click Verify again. Each round continues the same conversation, so your original inputs stay exactly as you entered them.
3. Once the model reports it is ready, an Accept button appears. Accepting writes the file and shows the cost of generating it.

Leaving a Forge page mid-way asks for confirmation, since an unaccepted draft is not saved anywhere.

PromptForge — UI prompts from screenshots

ui.md tells the agent what your app looks like and how it behaves — see UI information. Instead of writing it, open a ui.md in the Tests editor and click Forge from screenshots.

• Upload screenshots of the app (PNG or JPEG, up to 30) and write a short bullet-point description of what they show (at least 160 characters). Verify stays disabled until both are there.
• Run the Verify loop above.
• Click Accept & Update ui.md to write the section, or Add another UI section to describe another screen first — the finished sections wait until you accept them all together.

All sections are appended to the ui.md you opened Forge from (a fresh, still empty scaffold is replaced), and the editor reopens on that file. Read the result and refine it: ui.md steers every run.

ToolForge — custom tools from a description

A custom tool is a C# file the agent can call during a run. If you would rather not write the C# yourself, click Forge a tool on Extending → Custom Tools.

• Write a structured description: what the tool does, which input fields it takes (name and type), what it returns, and which device it needs (desktop, web, Android, iOS, or none).
• Optionally point it at an Example source folder — working code that already does what the tool should do. ToolForge reads those files and then asks only about the tool's contract and configuration, not about logic the code already shows.
• Run the Verify loop above.
• Click Accept & Forge Tool. The .cs file is written into utils/custom_tools/ once it compiles, and it joins the agent's toolbox on the next run.

Not everything is a tool. A custom tool is code the agent calls during a run. It can't install software on the machine, change system or network settings, or duplicate a built-in tool from the Tool Store. If your request falls outside those boundaries, generation is rejected and you're pointed to an AskUI solution engineer (if a built-in tool already covers it, you're told which one). The same happens if the generated code doesn't compile after a few attempts — nothing is written. Either way your description and answers are preserved and copyable, so you can send them to support@askui.com without retyping them.

TestForge — tests from recordings (coming soon)

The third Forge turns material you already have into tests: upload a narrated screencast of someone walking through a flow, or a written spec as PDF or CSV, and TestForge drafts one askui test per file through the same Verify loop. Folders and zip archives turn into a batch of tests that keep their file and folder names.

TestForge is in development and not yet in the app. This page will be updated when it ships.

The Forge Model

Every Forge runs on the model you choose under Settings → Account → Forge Model. By default it is the same model as your test runs.

• On the AskUI hub you can instead pick the best available model or any other hosted model. Forge runs only occasionally, so a stronger model is usually worth it.
• If you bring your own model, untick Same as for my test runs to configure a separate provider for Forge (endpoint, model, API key). See Model providers.

Every Verify and every accepted generation shows its estimated cost and token count; with your own model the dollar amount is omitted unless you have entered its rates.