Floot AI in your app
The AI service your app calls for its own users: what it does, the models, how it spends credits, how to switch it off, and using your own provider key.
Floot AI is the AI your app uses for its own features: a chatbot, a summarizer, an image generator. It needs no provider account and no API key. It is not the assistant you build with — Claude or ChatGPT builds the app, and Floot AI is what the finished app calls when people use it.
What it can do
- Text — generation, summaries, classification, and extraction, with tool calling and web search
- Images — generation and editing
- Embeddings — for semantic search over your own content
- Transcription — speech to text
Chatbots and agents are built on the same service, with conversation history and memory kept in your app's database — see What's built in.
Models
Your app's code names the model for each call. Three chat models are offered:
| Model | Use it for |
|---|---|
glm-5 (default) | Hard reasoning, coding, and agents that work through many steps. It always thinks before it answers, so one call takes tens of seconds to minutes. Text only. |
gpt-6-luna | The fast, low-cost choice: summaries, drafts, classification, short replies, and anything a person is waiting on. Reads images and PDFs and can search the web. |
gpt-6-sol | The most capable and the most expensive. For the hardest tasks, and for built-in web search, a code interpreter, and image generation inside a chat. |
One app can mix them: a fast model for what a person waits on, glm-5 for work in the background. Ask your assistant which model a feature uses and why, and ask it to switch if answers are too slow or too shallow. For a model that is not in this table, use your own provider key, below.
What it costs
Floot AI spends credits, the same credits image generation spends. They come from the project owner's balance, whoever is using the app, and calls made while you test in the preview count too. Each response reports the credits it used, so your app can show or record its own usage.
Services → AI in your project shows that project's total and a breakdown. Your account's balance is on the plan and usage page. For what each plan includes and how to buy more, see Billing, credits, and plans.
Your users spend your credits
Floot AI has no per-user or per-project limit of its own. Every call draws on your balance until it is empty, including while you are not building. When it runs out, AI calls fail with an out-of-credits error until you add credits. A free account's 3,000 credits are granted once, at signup, and do not refill. If the app is open to the public, ask your assistant to put AI features behind sign-in and to add a limit per user.
Switching it off for a project
Open Services → AI and turn the Floot AI switch off. It applies to that project only, and it takes effect for the preview and the published app without a publish. While it is off, every Floot AI call from that project is refused, so an AI feature that has not been moved to another provider stops working. Turn the switch back on to undo it.
Using your own provider key
If you would rather pay a provider directly, or need a model Floot AI does not offer, tell your assistant you want to use your own OpenAI, Anthropic, or other provider key instead of Floot AI. You enter the key through the secure link it gives you — see Connecting external services — and your app then calls the provider itself. Those calls spend no Floot credits; you pay the provider.
Things to know
- It does not learn from your app
- Each call stands alone. The model knows only what your app sends with the request, so a feature that should answer from your data has to pass that data in. That is what gets built when you ask for a chatbot that knows your content.
- Some answers take a while
glm-5can take a minute or more, and a call that uses tools such as web search takes longer than one that does not. That is the model working, not a fault. For something a person waits on, ask your assistant to usegpt-6-luna, or to stream the answer as it is written.