The same support brain, now on your website
RAUM Chat answers your customers before they ever open a ticket. It reads the same procedures and knowledge your ticket automation already uses, thinks before it replies, and asks for a human the moment it should.
This is the website channel. See the shared platform
Write it once. It answers everywhere.
RAUM Chat is not a second product with a second knowledge base to maintain. It reads the exact same procedures, guidelines, articles and notes your Zendesk automation reads, and runs them through the same four-step pipeline. Update a refund policy once, and both the ticket queue and your website start using it.
It researches before it answers
Most website chatbots pattern-match your help centre and hope. Every RAUM Chat reply runs the same four steps your tickets do.
Understand
Reads the question and the conversation so far, and decides what it still needs to know before answering.
Plan
Searches your procedures, guidelines and knowledge base, then plans how to answer this specific person.
Act
Looks up live data through the connections you have approved for public use. An action that changes something runs only if you have configured it for public use, and only after the visitor confirms it.
Reply
Streams the answer word by word, in the visitor's language, grounded in what it actually found.
Built for real conversations
Survives a refresh
Answers stream token by token and are saved as they are produced. If the connection drops or the visitor reloads mid-sentence, the conversation resumes exactly where it stopped instead of starting over.
Screenshots, both directions
Visitors can attach a screenshot (PNG or JPEG) when words are not enough, and the assistant can send back images from your own gallery to show rather than describe.
Hands off to a person
When someone asks for a human, the request goes to your inbox, a signed webhook, or straight into Zendesk as a ticket with the transcript attached — after the visitor explicitly consents to sharing it.
Your AI key, your bill
Connect your own OpenAI, Anthropic or Google account. You choose the model, you pay your provider directly, and we never mark it up.
Collects details, routes the rest
Ask for contact details before a chat starts, and show a handoff form when the visitor asks for a person, when the answer is low-confidence, when an action fails, when an intent you choose comes up, or outside the business hours you set. Fields can be translated, and pre-chat answers carry over to the handoff form where the fields match.
Verified visitor context
Your site can pass a signed statement of who a signed-in visitor is, or a visitor can verify their email address. Lookups and actions you configure can require that level of identity first. Anything a visitor merely types stays unverified.
Rehearse before it goes live
Test a widget privately with its own knowledge, languages and public policies before any visitor sees it. Proposed actions are recorded but never executed. This is separate from Simulation Mode for Zendesk tickets.
Rated, measured, reported
Visitors can rate a conversation and email themselves the transcript. A dedicated chatbot dashboard tracks conversation volume, containment and handoffs, CSAT, action outcomes, languages, topics and sentiment, per-widget comparison and AI cost.
One tag, and it is live
Create a channel, list the domains it is allowed to run on, and paste the snippet into your site. The widget loads lazily, so it costs nothing until someone opens it.
<script
src="https://app.raum.am/widget/v1/loader.js"
data-account="YOUR_CHANNEL_ID"
data-key="YOUR_PUBLISHABLE_KEY"
></script>Your channel ID and publishable key are generated when you create the channel, and can be rotated at any time.
Safe to point at the open internet
A public chatbot talks to anyone who visits, including people trying to make it misbehave. These limits are enforced in the system itself, not asked for in a prompt.
Actions only where you allow them
Lookups use only the connections you have approved for public use. Anything that changes state — a refund, a cancellation, updating a record — is off by default: it is offered to a visitor only if you have configured that action for public use, and it runs only after the visitor confirms it. These limits are checked on the server and every confirmed action is recorded; none of it depends on a prompt instruction.
Runs only where you say
The widget answers only on the exact domains you list. Publishable keys can be rotated without downtime, and rate limiting stores a hash of a visitor's address rather than the address itself.
You decide what is kept
Set how long transcripts live, export them, put one on legal hold, or schedule deletion. Deleted conversations leave a record that it happened and nothing of what was said.
Pricing
You are charged once per conversation, not per message and not per “resolution” — so nobody has to argue about what counts as resolved. Bring your own AI key and pay your provider directly, with no markup from us.
- 1 website channel(s)
- Handoff by email
- Your own AI key — no markup
- 3 website channel(s)
- Handoff by email, webhook or Zendesk
- Visitor image upload (PNG, JPEG)
- Your own AI key — no markup
- Unlimited website channels
- Handoff by email, webhook or Zendesk
- Visitor image upload (PNG, JPEG)
- Retention controls, legal hold, export
- Your own AI key — no markup
- Unlimited website channels
- Handoff by email, webhook or Zendesk
- Visitor image upload (PNG, JPEG)
- Retention controls, legal hold, export
- Your own AI key — no markup
RAUM Chat is set up for each customer by our team, so your knowledge base and connections are right before it talks to anyone. Get in touch and we will size a plan with you.
Interested in RAUM Chat?
RAUM Chat is currently set up for each customer by our team, so we can get your knowledge base and connections right before it starts talking to anyone. Tell us about your site and we will take it from there.
Get in touch