Your docs already have the answer. airys makes sure your users find it — automatically. You only touch the conversations that genuinely need you.
Join the private beta and get 3 months of the Team plan free.
This widget mirrors the same shape your customers will see. Tap a starter below — or type anything you like straight into the text box and hit Send.
Write your articles in the dashboard. Drop the widget in your app. Let airys answer the rest.
A clean CMS for support articles — rich text, collections, drafts, publish. Articles are auto-embedded into a vector index the moment you save them.
One script tag, anywhere on your site. Configure the look, the position, and which pages it appears on. No SDK gymnastics.
When the AI isn't confident, it opens a ticket — with the transcript, citations, and self-rated confidence attached. It follows your SLA, sends idle-ticket follow-ups, and extracts the resolution back into the knowledge base when the conversation wraps up. You only touch the conversations that actually need you.
Every time a conversation gets resolved, airys extracts the answer and adds it to the knowledge base. No retraining. No manual updates. The AI handling conversation #500 is meaningfully better than the one that handled #1 — because it learned from every resolution in between.
Every reply cites the exact help article it came from, so users (and your team) can verify what airys said. If we can't ground it, the AI says so — never invents an answer or a citation.
Every reply carries a self-rated confidence. When the AI can't ground an answer, it doesn't guess — it leaves a private note for your team and holds the customer with a brief reply. Only when something genuinely needs a person does it pull in the watching supervisor or open a ticket — always with the transcript, citations, and confidence score attached. No cold handoffs. No context lost.
The AI doesn't just hand off — it stays accountable. It tracks first-response, idle, and resolution thresholds per ticket priority. When a ticket nears its first-response target, it notifies the assigned agent before they breach.
See which questions airys couldn't answer. Turn them into help articles in one click.
Your articles get a branded, searchable site at help.yourdomain.com. Same content powers the widget — write once, surface everywhere.
Configure your primary color, accent, position, and greeting. The widget renders in an isolated shadow DOM so your site's CSS never bleeds in — but it always looks like part of your app.
Run your help center at help.yourdomain.com with automatic TLS. No DNS gymnastics, no third-party branding once you upgrade.
Not everyone opens a chat widget. Turn on the email channel and customers can just reply to an email — airys answers in-thread with the same grounded, cited replies, and your team can jump in anytime. On Solo and Team, send and receive from your own domain, like support@acme.com.
The Lab is a private sandbox to chat with your agent and check its answers — test chats never touch your analytics, usage, or what it learns. Tune the help center until the answers are right, then ship.
Instead of dumping the customer into a queue, the AI leaves an internal note for your team and replies with a brief holding line. Your answer in the note becomes its next reply — in its voice. The customer never sees a handoff happen, just the answer they were waiting for.
Deflect the easy 80% — and let the agent handle the rest. Onboarding questions, billing FAQs, and escalations all flow through one agent that follows your SLAs, learns from your team's replies, and gets better every week.
Write your API reference as articles — the widget pulls the right snippet on demand, with citations linking back to the source.
One flat dashboard for your help center, widget, and ticket inbox. When you do need to reply, the transcript and context are already there — no digging.
Every plan includes the full widget, the dashboard, and source-grounded answers. Your plan size is set by conversation volume, not headcount.
Sign up, write a few articles in the dashboard, customize the widget, and paste a script tag on your site. Articles are indexed the moment you publish — there's no separate "training" step.
Every answer is grounded in your indexed content — if airys can't find a source, it doesn't guess. It either asks a clarifying question or hands off to a human. We surface every citation back to your team so you can audit.
Conversation and ticket data is stored in MongoDB Atlas on AWS. Customer data is never used to train the underlying model. You can export or delete your project's data at any time.
No — that's the point. airys includes its own ticket inbox. When the AI isn't confident, it opens a ticket for that issue inside the conversation — with priority, SLA tracking, and automatic follow-up reminders. The conversation and the ticket are separate things.
One end-to-end chat session in the widget, regardless of how many back-and-forth messages it contains. Dashboard usage, article reads, and ticket replies inside the inbox don't count toward your limit.
A conversation is one end-to-end widget session — that's the billing unit. Within a conversation, the AI may open one or more tickets for the things it can't resolve. A customer who asks three questions — two answered by the AI, one escalated — counts as one conversation, with one ticket in your inbox. You're never charged per escalation. Some asks become feature requests instead of tickets — the AI routes those to a separate roadmap board so the product team sees demand build up.
The hero on this page has a live airys widget — ask it anything. For a tailored demo with your own docs, request access below and we'll get you onboarded.
We're onboarding new teams every Tuesday. Get on the list and we'll reach out when your slot opens — beta teams get 3 months of the Team plan free.
Or email hello@airys.ai directly.