← All releases
1.1.0 OpenAI and DeepSeek
- New: choose between Anthropic (Claude), OpenAI (GPT) and DeepSeek right from the AI Provider tab, each with its own selection of models — from the fastest and cheapest for high chat volume, up to the most capable
- DeepSeek in particular is dramatically cheaper than the alternatives, which makes a busy chat widget far more affordable to run
- Every provider keeps its own API key and chosen model saved separately, so you can try one, switch back, and pick up exactly where you left off
- The live cost estimate and the daily spending cap work the same for all three, so you always know what a reply costs before you commit to it
- Contact details the assistant gives out — a phone number, a support email, a social profile, any web address — now appear as tappable buttons instead of plain text the visitor had to select and copy by hand. On a phone that means one tap to call or to write, at exactly the moment someone decided to get in touch
- Fixed: when a conversation hit its maximum length, the message asked visitors to start a new conversation — but there was no way for them to do that. It now explains that they can continue after a short wait (if you had customized this message yourself, your own wording is kept)
- Fixed: on the rare occasion the AI returned an empty reply, the widget went on showing the typing dots forever, with nothing to tell the visitor it had finished. It now shows a short message instead, which you can reword under Widget Text
- Fixed: the cost estimate was too optimistic. It assumed a small, fixed amount of text per message instead of measuring the business information you have actually filled in — which is re-sent to the AI on every single message. It now reflects your real setup, so the number you plan with is the number you get