Quest Ladder
TeachFORGE.ai
Quest Ladder / Where the AI runs
Where the AI runs: copy, your GPU, your key
The same prompt, three routes. Only the model and the bill change.
Same prompt everywhere. The difference is the model.
The three modes
| Mode | What it does | Who pays |
|---|---|---|
| Copy prompt | Build the prompt here, run it in any chat model, paste the JSON reply back. | Nobody — free on every plan |
| My GPU | One click to the endpoint you configured under Profile → My AI Endpoints. | Nobody — it is your hardware |
| My API key | One click using your own OpenAI or Anthropic key. | Your provider, directly |
TeachForge sells no AI credits and runs no AI for you, so these three are the whole list. See AI credits: there are none to buy.
The same prompt everywhere
The panel states it plainly: every mode uses the same compiled prompt and the same task defaults, and only the model and the billing differ.
That is worth internalising, because it settles a question people spend time on. If your own key gives you a better type than your own GPU, the cause is the model you are pointing at, not the platform doing something different. Try a stronger model on your own endpoint before concluding otherwise.
What each needs
- Copy prompt — nothing. It is the fallback that always works, and free on Free.
- My GPU — an endpoint configured on your profile. See Connect your own GPU.
- My API key — a provider key saved on your profile. See Use your own key.
Both bring-your-own routes are free of TeachForge billing entirely. They need a Standard plan or higher, and an administrator must have enabled them for the site.
While it runs
A run survives navigation — you can work elsewhere in TeachForge while it finishes — but closing the tab ends it, and your own GPU specifically needs the tab kept open because the browser is what talks to it.
Only one generation runs at a time; starting a second says so. Stop cancels the current one.
When a reply does not parse, the exact text the model returned is available to read, and Paste JSON manually lets you correct and submit it rather than losing the run.