AI chatbot
Ask in plain English.
Watch the AI work.
A grounded chatbot for your knowledge base, with an optional agentic search loop (Business and Enterprise) that keeps trying when the first query doesn't land. Every answer cited, no hallucinations, fully optional.
Search works when you know the words. What about when you don't?
Most AI chatbots are bolted onto knowledge bases as an afterthought. They're either too willing to invent answers from training data, or too literal to bridge the gap between how a customer asks a question and how your team wrote the article.
A chatbot that hallucinates is worse than no chatbot, because every wrong answer chips away at trust. A chatbot that gives up at the first weak match might as well be a dumb FAQ.
You need a chatbot that's grounded, persistent, and honest about what it doesn't know.
Grounded answers, with the AI's reasoning visible.
The chatbot reads from your knowledge base via retrieval-augmented generation. A user asks a question, the chatbot searches your articles, documents, and slides, and returns an answer with citations to the exact sources. If the answer isn't in your content, it says so rather than making something up.
The agentic search loop (an opt-in toggle on Business and Enterprise) kicks in when the first internal search returns weak results. The AI rewrites the query — using different words, related concepts, alternate phrasings — and tries again. Each rewritten search is visible live, so the user can see what the AI is trying. They can stop early, refine manually, or let it run. The loop costs more AI credits per question; admins switch it on when the search-quality lift is worth the spend.
When the chatbot can't find an answer, the configurable escalation action takes over: a contact form that emails you, or a URL link to your support flow. Customers and team members always have a clear next step.
One AI chatbot. Four places to deploy it.
Same grounded AI. Same citations. Same audit trail. Different audience, scope, and content visibility per surface.
Team-facing
In-app for your team
Inside the KnowledgeScout app. Scoped to whatever your team is allowed to see based on their role and team-scoped article access. Mid-call agents, new starters, anyone needing a fast answer can ask in plain English.
Available on every plan.
Customer-facing
Branded help centre
Public portal on your custom domain (help.yourcompany.com) or your-company.knowledgescout.io. Customers ask in plain English, get cited answers from your public articles. Internal content stays internal regardless of how it's prompted.
Business and Enterprise.
Team-facing
Internal embeddable widget
Drop the same in-app chatbot into your CRM, intranet, or any internal tool. Same role and team scope. Your team gets the chatbot where they already work, without context-switching.
Business and Enterprise.
Customer-facing
External help widget
Floating chat widget you embed on any page of your customer-facing site. Same public-only scope as the help centre. Loads async, inherits your colours, no full-page redirect for your customers.
Business and Enterprise.
You don't have to deploy all four. Drop the customer widget on your existing site and skip the help centre. Embed the internal widget in your CRM and skip the in-app surface. Each one stands alone. Pick the surfaces that fit how your team and customers already work.
What makes it different
Most AI chatbots are an opaque box that returns a confident-sounding answer. Ours shows its working.
Agentic search loop (Business + Enterprise)
When the first internal search returns weak results, the AI rewrites the query and tries again. The loop keeps going until something useful turns up or the chatbot decides there's no answer to give. Opt-in toggle, uses additional AI credits per question.
Live reasoning trail
Every rewritten search is visible to the user as it happens. They can see what's being tried, stop the loop, or refine manually. No black box.
Citations on every answer
Every claim links to the source article, document, or slide. Users verify in one click. AI agents reading via MCP get the same citations.
No hallucinations
The chatbot only answers from your knowledge base. If the answer isn't there, it says so. Your content is never used to train the underlying model.
Configurable escalation
When the chatbot can't find an answer, the escalation action is yours to set: a contact form or a URL link to your support flow. Users always have a clear next step.
Bring your own AI key
On Business and Enterprise, connect your own Anthropic, OpenAI, Azure, or OpenAI-compatible key. Unlimited usage at your provider's rates, with no double-pay through us.
Permission-scoped
Internal chatbot respects role and team-scoped article access. Customer-facing chatbot only ever sees public articles, no matter how it's prompted.
Optional, always
Opt-in per workspace. Turn the chatbot on for the team only, customers only, or not at all. The knowledge base works completely without it.
Chat analytics
See what's being asked, where the chatbot didn't have an answer, and which queries triggered escalation. Direct feedback on what content to write next. More on analytics →
Configurable conversation log
Don't store conversation content at all, or store full transcripts up to seven years for compliance. Configurable per tenant, separately for internal team chat and customer-facing chat.
Who this is for
Teams whose users don't know your jargon
Your articles call it "Annual Leave Policy v3." Your customer asks about "PTO." The agentic loop bridges the gap so the user finds the right article without learning your taxonomy first.
Contact centres handling fuzzy questions
When a customer asks "what's the deal with refunds again," your agent doesn't have time to translate. The chatbot does the rephrasing live, mid-call, with the right article surfaced and cited.
Customer support replacing tier-1 human triage
If you're still answering "where do I update my billing details" by hand, that's the chatbot's job. Configure the escalation for the harder questions and let your team focus on those.
Regulated industries that can't tolerate hallucinations
Citations on every answer, audit trail on every conversation, no training on your data, no invented policies. The chatbot you can put in front of customers without legal flinching.
Why KnowledgeScout's chatbot
Grounded in your content, never the internet
The chatbot reads from your knowledge base only. It doesn't browse the web, doesn't pull from training data, and doesn't combine sources outside your tenant. Every answer is traceable to something a human at your company wrote.
The reasoning is visible
Most AI chatbots run their internal queries silently and present a single confident answer. Ours shows each rewritten search as the agentic loop runs. Users learn what works, you can audit what was tried.
Optional, with no penalty for opting out
If your team can't or won't use AI, the knowledge base is still fully usable. No greyed-out features. No nag prompts. Just a knowledge base.
BYO key means no AI markup
On Business and Enterprise, connect your own AI provider key. You pay your provider directly at their rates, not us at a marked-up rate. Useful when you have an existing AI agreement or want a specific model.
Common questions
How does the agentic search loop work?
The agentic search loop is an opt-in toggle on Business and Enterprise — off by default. When enabled, weak first-pass results trigger an AI rewrite of the query (handling synonyms, alternate phrasing, related concepts) and a retry. The loop continues until the chatbot finds something useful or decides the answer isn't in the knowledge base. Every query the AI generates is shown live to the user, so they can see what's being tried, stop the loop early, or refine manually. Stronger search results, in exchange for more AI credits per question.
Can I turn the chatbot off entirely?
Yes. The AI Chatbot is opt-in per workspace. If your organisation has AI restrictions or your team prefers no AI in the loop, leave it off. The knowledge base works completely without it — search alone is a complete way to use the platform.
Does the chatbot hallucinate?
The chatbot is grounded in your knowledge base via retrieval-augmented generation. It only answers from articles, documents, and slides that exist in your tenant, with citations to the source for every claim. If the answer isn't in your content, it says so rather than inventing one. Your content is never used to train the underlying model.
Which plans include the chatbot?
Every plan, including Startup, includes the AI Chatbot with included AI credits. The agentic search loop (the rewrite-and-retry feature for weak queries) is on Business and Enterprise only, and is an opt-in toggle. Business and Enterprise also add bring-your-own AI key support, which means unlimited usage at the cost your AI provider charges you directly.
Can I bring my own AI provider key?
Yes, on Business and Enterprise plans. Connect your own Anthropic, OpenAI, Microsoft Azure OpenAI, or any OpenAI-compatible API key (including self-hosted models). Usage is billed by your provider directly. Useful when you have an existing AI agreement, prefer a specific model, or want to avoid double-paying for AI.
Can the chatbot answer customer questions, not just internal ones?
Yes. The customer-facing chatbot runs on two surfaces: the branded help centre and the embeddable widget you can drop on any page of your customer-facing site. Both are on Business and Enterprise, both are scoped to public articles only. Customers ask in plain English, get cited answers, and a configurable escalation path handles questions the chatbot can't answer. Internal articles never leak to customers, regardless of which surface they ask through.
Do I need a help centre to use the AI chatbot?
No. The four surfaces are independent — pick the ones that fit. If you want AI chat on your customer-facing site, drop the external widget on any page without standing up a help centre. If you want it for your team, the in-app chatbot and the internal embeddable widget work without you exposing anything externally. The help centre is one of four surfaces, not a prerequisite for any of the others.
A chatbot that thinks, not one that guesses.
Grounded answers and citations on every plan. Agentic search loop on Business and Enterprise. Optional in every workspace.