v20 — Widget 2.0: a smarter launcher, voice input, and a faster, safer chat

v20 is a ground-up upgrade of the on-site chat widget — the part your visitors actually touch. It now greets people with a configurable peek bubble, accepts voice input, streams answers word-by-word with proper Markdown and source attribution, and loads faster while being noticeably harder to abuse. Every setting below is per-bot in the admin under Behavior.

👋 A smarter launcher

• Peek greeting bubble — a friendly one-line greeting pops up beside the launcher a couple of seconds after the page settles, on every visit. Fully configurable: turn it on/off, set custom text (or let it reuse your welcome message), and tune the delay (default 2000ms).
• Auto-open and the peek now work together — the peek can tease first, then the panel auto-opens and clears it cleanly (just set the peek delay shorter than the auto-open delay).
• Unread badge — if a reply lands while the panel is closed, the launcher shows a count, and the idle "online" dot steps aside so the badge stays visible.

🎙️ Voice input

Visitors can talk to the bot instead of typing.

• Free plan uses the visitor's browser speech recognition — instant, on-device, no added cost.
• Paid plans add OpenAI transcription (Whisper) as an automatic fallback, so voice keeps working even on networks or browsers where the built-in speech service fails.
• Clear, friendly prompts when the mic is blocked, the network drops, or nothing was heard — no more silent failures.

💬 Better answers

• Markdown rendering — bullet lists, headings, bold, and links render properly in replies (safely escaped first).
• Source attribution — answers grounded in your knowledge show a subtle "sourced from" line.
• Word-by-word streaming with a real typing indicator that cycles through "Searching knowledge…" then "Drafting reply…".
• Quick-reply chips — up to three suggested follow-ups after each answer.
• "Show more" on long answers — continue a truncated reply with one tap.

⚡ Faster & safer

• Async, lighter embed that never blocks the host page, with stronger style isolation so the widget looks identical on any site.
• Smarter rate-limiting — per-bot ceilings with a graceful "I'm getting a lot of questions right now" message instead of hard errors, plus accurate per-visitor limits behind our edge.
• Tighter privacy — the public widget config no longer exposes internal bot configuration; answers still use your full knowledge server-side.
• A message-length cap to keep things tidy and costs predictable.

💳 Billing

onlyAI now bills through Lemon Squeezy as Merchant of Record (EU VAT and sales tax handled for you), with a choice of monthly or annual plans.

It's all live now — open any bot in the admin, head to Behavior, and tune the launcher to taste.