AI that works as your customer service agent, 24/7.
It’s ChatGPT built just for your product. Trained on your content, answering 24/7, with every reply citing the exact page it came from.
- Trained on your own content
- Every answer cites its source
- 95+ languages
One agent, shaped to how you sell.
Pick your business to see the questions it answers and the leads it captures.
Most site chatbots guess. Then they lose the customer.
A wrong answer given with confidence is worse than no answer at all.
Buyers ask specific things: which plan has SSO, what your refund window is, whether you work with their stack. A model that never saw your content answers from the average of the internet, and the average is not your product.
It invents an answer
With none of your content to draw on, it fills the gap with a confident guess. One wrong answer about your plans and the buyer stops trusting every reply after it.
It cannot show its work
An answer you cannot check is one you cannot act on. If the bot will not link the page it used, your team cannot see why it said what it said.
It lets the lead go
A visitor ready to buy at midnight hits a dead end. No question captured, no detail asked, nothing waiting for your team in the morning.
Answers from your content, with the receipt.
What it does on every conversation, unattended.
From a URL to a working support agent.
No reformatting, no per-answer scripting. You point, it trains, you deploy.
- 1
Point it at your content
Paste a URL or sitemap, or upload PDFs, DOCX and text. It reads them directly.
- 2
It trains itself
It chunks, embeds and indexes your content in minutes. Watch progress in the dashboard.
- 3
Set the boundaries
Turn on strict mode, pick the lead fields to collect, and set the handoff rule.
- 4
Deploy anywhere
Drop one snippet on your site, or connect a channel. The same agent answers in each.
<script src="…/widget/loader.js" data-bot-id="your-bot-id"></script>Install guide How an answer gets made.
Two model calls, named so you know exactly what touches your data.
Where does my data go?
Your content is chunked and stored as vectors in your own workspace. It is used to answer your visitors and nothing else.
What writes the answer?
Claude drafts the reply from the passages retrieved from your content. OpenAI embeddings power the retrieval step.
What if there is no source?
In strict mode the agent says it does not have that answer and offers a person, instead of guessing.
Does it stay current?
It re-crawls your sources on a schedule that scales with your plan, so answers follow your documentation.
- Your contentSite, sitemap, files, connectors
- Chunked and embeddedOpenAI embeddings
- Stored in your workspacepgvector, row-level security
- Top passages retrievedFor each question
- Claude drafts the replyFrom those passages only
- Answer with citationUp to 4 source pills
Built for the people who answer buyers.
The same trained agent covers the three jobs that eat a support and sales team’s day, so a person only steps in when it counts.
- Support deflectionIt answers the same billing, setup and policy questions all day so your team handles the ones that need a person.
- Lead captureIt qualifies and records interested visitors at 2am and hands your team a lead in the morning.
- Meeting bookingIt answers the pre-sales question, then offers a “Pick a time” button once, while the reason to book is still fresh.
Learns from where your content lives. Answers where your customers are.
Sources
Channels
Sources: Website crawl, XML sitemap, PDF and DOCX, Pasted text, Notion, Google Drive, Dropbox, OneDrive, SharePoint, Box, Confluence, GitBook, Zendesk, Freshdesk, YouTube. Channels: Website widget, Slack, Messenger, Google Chat, Zoho, Crisp, Freshchat, Zendesk, WhatsApp, HubSpot.
The bot on this page runs on Matram.
No staged screenshots. The assistant in the corner is a Matram agent trained on Matram’s own docs. Ask it a hard question, then click the source it links and check it against the page. That round trip is the whole pitch.
- Every answer links the source page it used
- Answers in 95+ languages with no per-locale setup
- Re-syncs your content on a schedule
- Escalates to your inbox with the transcript attached
- Free 30-day free trial, no credit card
Honest guides from the team.
How AI support actually works, what it costs, and how to choose.
Chatbot resolution rate vs escalation rate: which numbers to trackResolution rate measures conversations completed without human help. Escalation rate measures conversations passed to a person.
After hours lead capture: what happens to your website visitors at 2amAfter hours lead capture should do more than collect an email while your team is offline. Answer the visitor’s question before asking for details. Colle...
How to Reduce Demo No-Shows When an AI Agent Books the MeetingShort answer To reduce demo no shows, do more than place a calendar link in the chat. Use Matram to answer the visitor’s questions, check whether a demo...
Questions, answered.
Still deciding? Point it at your site and ask it your hardest support question on the free trial.
What is Matram?
Matram is an AI support agent trained on your own content. It answers customer questions from your site, docs and help center, links the source page for each answer, captures leads, and hands off to a person when it should.
How does it avoid making things up?
It answers from passages of your own content and cites the page it used. Strict mode restricts it to approved material, so it says it does not know rather than guessing.
How long does setup take?
Minutes. Paste a URL or sitemap, or upload files. Matram chunks, embeds and indexes the content, then you test it and drop one snippet on your site.
Which channels does it work in?
The website widget today, with Slack, Messenger, Google Chat, Zoho and WhatsApp as you connect them. The same trained agent answers in every one.
What does it cost?
Plans start at $29 a month with a 30-day free trial and no card required. The pricing page lists what each plan includes.
Can it hand off to a human?
Yes. It passes the conversation to your shared inbox with the transcript attached, and your team replies in the channel the visitor is already using.





