We think support bots should actually read the docs
JestBot started as an internal tool for answering the same three questions on repeat — and turned into a platform anyone can point at their own content.
No card required · Free plan included
Test Chat
Retrieved in 380msDo you support SSO?
Yes — SAML and OIDC are on the Team plan and above. Turn it on under Settings → Security.
An internal tool
Started as
An open platform
Became
Grounded > clever
Philosophy
AI, then human
Support model
How we got here
The problem
The same three questions, on repeat
JestBot started as an internal tool for answering the same handful of support questions over and over — the docs already had the answer, nobody was reading them.
The shift
From FAQ page to something that reads it for you
Instead of writing another FAQ page, we built retrieval on top of the docs directly — so the answer stayed correct even as the underlying content changed.
Today
A platform anyone can point at their own content
What started as an internal tool is now a platform: bring your docs, pick a model or use the free default, and it's answering questions on your site the same day — with a human safety net behind it.
“We built the first version because I was answering the same billing question for the fourth time that week. It wasn't glamorous — it just worked, so we kept it.”
— From the founding notes
How we think about it
A support bot should read the docs before it speaks
Three ideas we keep coming back to when we decide what to build next.
Grounded beats clever
A confident wrong answer is worse than no answer. Every reply JestBot gives is built from the content you actually provided — retrieval first, generation second — instead of a model improvising from general training data.
- Retrieval before generation
- Sources visible, not hidden
No black box
You should be able to see exactly why a bot said what it said. Test Chat shows the retrieved chunks and their similarity score for every response, so a bad answer is something you can debug, not just distrust.
- Retrieved chunks shown per reply
- Debuggable, not opaque
Humans are the fallback, not an afterthought
AI should take the first shot at every question — but the moment a conversation needs judgment, empathy, or an exception to a policy, a real person should be one click away, with full context, not a cold handoff.
- Handover built in from day one
- No context lost at handoff
A short list, kept short on purpose
Your data stays yours
Documents you upload power your own bot's retrieval only — never used to train a shared model.
Built with support teams, not just for them
Handover, business hours, and canned responses exist because real agents told us the AI-only flow wasn't enough on its own.
Ship the same day
No prompt-engineering sprint, no separate vector database to run — a bot going from zero to live on your site in one sitting.
What JestBot isn't
Easier to say what we optimize for than what we ruled out. Here's the second one.
Not a wrapper around a general model
JestBot doesn't forward your question to a generic model and hope for the best. It retrieves from your content first — if nothing relevant comes back, it says so instead of guessing.
Not a re-training step
Edit your docs and the index updates on its own. No export, no fine-tuning job, no answers that go stale three weeks after your content changes.
Not a dead end when the bot's stuck
"Contact us" isn't the fallback. A real person can take over mid-conversation with the full transcript already in front of them.