● RAG support chatbot
Your docs.
Now answering.
Train an AI support agent on your documentation and embed it on any site with one script tag. Every answer comes from your content and names its source. If the docs don't cover it, the bot says so.
demo@answerly.app / Demo123!
1 tag
to install on any site
6
chunks retrieved per answer
0
answers without a source
~2s
median response time
Support that knows your product
Feed it anything
PDFs, pasted text, or a URL, everything is chunked and embedded into a private vector index per bot.
Citations on every answer
Replies name the exact source documents they came from, so visitors (and you) can trust them.
Refuses to make things up
No relevant context, no answer, the bot says so instead of inventing a refund policy.
Matches your brand
Name, color, welcome and fallback messages, configured in the dashboard, live everywhere instantly.
Analytics that matter
Conversations per day, answered-vs-fallback rate, and the exact questions your docs failed to cover.
Locked-down public API
Origin allowlists, per-visitor and per-bot rate limits, and a kill switch, abuse never reaches your quota.
Live in three steps
$ answerly ingest ./docs
Feed it your docs
Upload files, paste text, or point it at a URL. Content is chunked along headings and embedded into pgvector.
$ answerly ask "How do I export?"
It answers with receipts
Questions retrieve the closest chunks; the model answers only from them and cites which sources it used.
$ <script src=".../embed.js">
Embed it anywhere
One script tag drops a branded chat bubble onto any website. Conversations stream into your dashboard.
<script src="https://answerly-liard.vercel.app/embed.js"
data-bot-id="your-bot-id" async></script>Fair questions
How is this different from pasting docs into ChatGPT?+
Answerly is a system, not a session: your knowledge is permanently indexed, answers cite their sources, off-topic questions are refused, and the whole thing lives on your website answering customers while you sleep.
What stops it from hallucinating?+
Retrieval-augmented generation with a hard gate: if no chunk of your content is similar enough to the question, the bot returns your fallback message without ever calling the model. When it does answer, it's constrained to the retrieved context and must cite it.
Can I control which sites can use my bot?+
Yes, set an origin allowlist and the public chat API rejects every other website. Bot ids are unguessable, and per-visitor rate limits stop abuse.
What happens when my docs change?+
Re-index the source with one click, old chunks are replaced atomically and the next answer uses the new content. Delete a source and its knowledge is unanswerable immediately.
Ship a support agent this afternoon
Upload your docs, copy one script tag, done.