How a ChatFrom chatbot behaves, its AI models and languages
Updated
Setting how a ChatFrom chatbot talks, the eight AI models it can use, the languages it speaks, and what it does when it cannot answer.
How do I set how the chatbot talks?
Under Behaviour you write the chatbot's instructions, which it reads before every answer (sometimes called the system prompt). You do not have to write them yourself: open Need help?, describe your business and what the chatbot should do, and press Write my setup. ChatFrom writes the instructions and fills in the settings below them. Nothing is saved until you have read it and pressed save. This uses a few credits, and it tells you how many first.
The settings under the instructions:
- Personality: how formal or friendly it sounds, and how much detail it gives.
- Language: the visitor's own language, or one you choose.
- When it does not know: what it says when the knowledge has no answer, and whether it then passes the conversation to your team.
- Guardrails: topics it politely declines, plus switches such as staying on topic, protecting its instructions, not asking for sensitive data, and not giving medical, legal or financial advice.
- Example answers: a few questions with the answers you would give, so it learns your style.
- Welcome and suggested questions: the first message visitors see, and a few questions they can tap.
Which AI models can a ChatFrom chatbot use?
Six models from four companies, each company's newest, chosen per chatbot and reached through OpenRouter:
- OpenAI: GPT-6 Luna (the default, fast and cheap), GPT-6.1 Sol (best quality).
- Anthropic: Claude Haiku 5.5 (fast and cheap), Claude Sonnet 5.5 (careful, long answers).
- Google: Gemini 3.5 Flash-Lite (fast and cheap).
- Mistral: Mistral Large 4, from a European provider.
The Model page shows how many credits an answer uses on each one and roughly how many answers a month your plan gives with it, so you can compare before switching. Larger models handle hard questions better but use more credits per answer.
The Model page also has a temperature, which makes wording more consistent or more varied, for the models that take one. GPT-6 Luna, GPT-6.1 Sol, Claude Haiku 5.5 and Claude Sonnet 5.5 set their own, so for them there is none to choose.
Which model should I choose?
Start with the default, GPT-6 Luna. For answering from your own documents, a small model is usually enough, because the hard part, finding the right passage, happens before the model is asked. Move to a larger model only if answers need more reasoning.
What happens if a model is unavailable?
If the chosen model cannot be used at that moment, for example because its provider stops serving it, visitors are answered with another model from the same provider, when it offers one, rather than turned away. Answers never come from a provider you did not choose. Small steps around an answer, such as turning a question into a search or reading a picture, use ChatFrom's own models, currently from OpenAI.
What happened to GPT-5 mini and the other earlier models?
On 16 October 2026 each was replaced by its own company's newest model, and a chatbot set to one now answers on that one, so its answers stay with the company you chose:
- GPT-5 mini and GPT-5 nano: GPT-6 Luna. GPT-5: GPT-6.1 Sol.
- Claude Haiku 4.5: Claude Haiku 5.5. Claude Sonnet 4.5: Claude Sonnet 5.5.
- Gemini 2.5 Flash and Gemini 2.5 Pro: Gemini 3.5 Flash-Lite.
- Mistral Medium 3: Mistral Large 4.
Nothing needs doing. The Model page shows the model your chatbot answers on, and you can pick another there at any time.
Is my data used to train models?
No. Nothing put into ChatFrom is used to train any model of ChatFrom's, and ChatFrom asks OpenRouter for providers that do not train on what passes through them. OpenRouter is told to use only the chosen model's own maker, never another company that also runs the model.
Which languages does the chatbot speak?
By default it answers in the language each visitor writes in, even when your documents are in another language. You can also set it to always answer in one of 14 languages: English, Swedish, German, French, Spanish, Italian, Dutch, Portuguese, Danish, Norwegian, Finnish, Polish, Turkish or Persian.
The dashboard is in English. ChatFrom's own support answers in English or German.
Does it say it is an AI?
Yes, by default. Under the chat, on your site and on the share page, visitors see "You're chatting with an AI assistant", and in chat apps the first answer of each conversation starts with that line. The EU AI Act says people must be told when they are talking to an AI unless it is already obvious, so the switch Tell visitors it's an AI assistant is on unless you turn it off.
What does the chatbot do when it cannot answer?
It says so, in the words you set, instead of making something up, and it can offer to hand the conversation to a person by email, Slack or a ticket in your own system. Visitors can rate every answer, and you can turn a weak answer into a Q&A entry with Fix this.