What is an AI knowledge base, exactly?

It's a store of your own documents — policies, product specs, FAQs — that an AI agent searches before it answers, so replies come from your real content instead of a general model's guess. KlyoChat chunks each file, stores it as pgvector embeddings, and retrieves the most relevant passages for every question. The technique is called retrieval-augmented generation, or RAG.

What file types can I train the chatbot on?

PDF, Word (DOCX), plain text, and images. KlyoChat chunks each file, generates pgvector embeddings, and makes it retrievable by your AI agents. You can mix a policy PDF, an FAQ doc, and a screenshot in the same knowledge base.

How does the AI actually answer from my documents?

For every question, the agent turns your docs and the customer's message into embeddings, ranks your chunks by meaning, and writes the reply using only the closest matches. Because it matches on meaning rather than exact keywords, a customer can phrase a question nothing like your document and still get the right passage.

Do scanned PDFs or image-only files work?

Images and screenshots are a supported source type, so spec sheets, menus, and screenshots are readable. For a PDF that is entirely scanned pages, retrieval depends on the text KlyoChat can extract — a file with real, selectable text always performs better than a photo of a page.

Will the AI make things up (hallucinate)?

Strict grounding limits answers to what's in your docs, and each reply carries a citation you can check. When there's no matching source — a policy or price it can't find — the agent escalates to a human instead of guessing. That restraint is the whole point.

What happens when the answer isn't in my documents?

With strict grounding on, the agent doesn't improvise. If nothing in your knowledge base matches, it hands the conversation to a human instead of inventing a return window or a price — you set that escalation rule once and it holds across every channel.

How do citations work, and can my team see the source?

Every grounded reply arrives with the exact chunk it was written from — the source file, the passage, and a relevance score. A teammate can confirm the agent quoted your real policy in one click, instead of trusting a paraphrase that only sounds right.

How is this different from ManyChat's AI Knowledge or Chatfuel?

Those tools bolt AI onto a flow builder and answer from a general prompt — you can't see where a reply came from. KlyoChat grounds every answer in your actual documents and attaches the source chunk, so you can verify it, and Chatfuel notably can't be deeply trained on your SOPs or catalog the same way. Feature sets and plan limits change often, so check each tool's current docs before deciding.

Can ManyChat or Chatfuel show me the source behind an answer?

Not the way KlyoChat does. Both are flow builders with an AI layer on top — ManyChat routes intents to pre-built paths, Chatfuel adds a general model — and neither attaches the passage a reply came from. That citation is the trust layer they don't surface, though you should compare current features and pricing directly, since all three ship changes regularly.

How does this compare to Intercom or Tidio's AI?

Intercom's Fin and Tidio's AI are strong products, but they're built around website chat and help centers. KlyoChat is DM-native — the same grounded, cited answers run across Instagram, WhatsApp, Telegram, Facebook, and X from one inbox. Pricing models differ a lot, so check current plans for the channels you actually use.

Which plan includes the AI knowledge base, and what does it cost?

The knowledge base is part of KlyoChat's AI agent capabilities. Plans and limits — document count, storage, and AI reply volume — do change, so see the pricing page for what's included today, and use the free trial to test it on your own docs first.

How long does it take to set up?

Most teams go from first upload to a live, cited answer in under ten minutes. You drop in a document, KlyoChat chunks and embeds it, and you point an agent at it — no data-science setup, no separate vector database to run, and no code.

How many documents can I add?

You can load multiple files into one knowledge base and mix formats freely. Exact document counts and storage depend on your plan and can shift over time, so check the current limits — but answer quality depends far more on clean, well-structured sources than on sheer volume.

Where is my data stored, and who can see it?

Your content lives as pgvector embeddings in Postgres — not a sealed box you can't reason about. It's used to answer your own customers' questions, and technical evaluators get a storage model they can actually audit.

How do I keep answers accurate when a policy changes?

Update the source file and the agent answers from the new version — there's no second copy to sync. Support teams tend to refresh knowledge-base content within a day of any real change, so the agent never quotes a stale price or return window.

Do I have to re-upload documents I already have?

No. Reuse files already sitting in your workspace by pointing an agent at them. There's no separate knowledge library to maintain, and updating the original file updates what the agent knows.

AI Knowledge Base

Turn your docs into answers customers can trust.

Upload a PDF, Word file, or screenshot and KlyoChat's AI agents answer from it word-for-word — then show the exact source behind every reply. No invented policies, no confident guessing.

Start free trial See pricing

knowledge · support-bot

1Ingest2Ask

Knowledge baseempty

Returns Policy.pdf

PDF · 12 KB · not ingested

Add a sample policy, then ask a question — watch the answer cite the exact chunk it came from.

Live now

What is AI Knowledge Base?

An AI knowledge base is a searchable store of your own documents that an AI agent reads before it answers, so replies come from your real policies and product facts instead of a general model's guess. In KlyoChat, you upload PDFs, Word files, text, or images; each file is chunked and stored as pgvector embeddings in Postgres, and your AI agents retrieve the most relevant passages to answer every DM. Every reply carries a citation back to the exact source chunk, so you can confirm the agent quoted your document rather than a plausible-sounding paraphrase. This retrieval-augmented (RAG) approach is what keeps answers grounded and safe across Instagram, WhatsApp, Telegram, and your other channels.

4 formats

PDF, Word, text, and images — ingested into one knowledge base

Every reply

carries a citation back to the exact source chunk

pgvector

embeddings stored in Postgres — your data, under your control

Results teams see

What AI Knowledge Base changes

70%

of routine questions answered from your docs

<10 min

from first upload to a cited answer

invented policies with strict grounding on

24/7

grounded answers with no one on shift

How it works

From upload to cited answer

Point an agent at your files, turn on strict grounding, and every reply comes back with its source. Here's the agent config and the answer it returns.

Ingest your docs

Drop in PDFs, Word files, text, or images. KlyoChat chunks each one and stores pgvector embeddings — no data-science setup on your side.

Or reuse what's in your workspace

Point an agent at files your team already uploaded. No parallel document library, no re-uploading the same policy twice.

Agents retrieve, then answer

For every question, the agent semantically ranks your chunks and writes the reply using only the most relevant ones.

Every reply cites its source

The answer arrives with the exact chunk behind it, so a teammate can verify in one click — grounded and on-brand.

1234567891011121314151617{  "agent": "support-bot",  "knowledge_base": {    "sources": [      { "file": "Returns Policy.pdf", "type": "pdf" },      { "file": "Shipping FAQ.docx", "type": "docx" },      { "workspace_file_id": "wf_8213", "type": "image" }    ],    "embeddings": { "store": "pgvector", "model": "text-embedding-3-large" },    "chunk": { "size": 800, "overlap": 120 }  },  "grounding": {    "mode": "strict",    "cite_sources": true,    "on_no_match": "escalate_to_human"  }}

What you get

Everything in AI Knowledge Base

Ingest PDF, Word, text — and images

Most tools stop at plain text. KlyoChat reads product PDFs, policy DOCX files, raw text, and images like spec sheets, menus, or screenshots — all into one retrieval-augmented knowledge base.

A citation on every answer

Each reply arrives with the exact source chunk attached, so a teammate can confirm the AI quoted the real policy — not a plausible-sounding paraphrase. This is the trust layer ManyChat and Chatfuel don't show.

Grounded, not guessing

Strict mode answers only from your documents. When nothing matches, the agent escalates to a human instead of inventing a return window or a price — the difference between helpful and confidently wrong.

pgvector embeddings you control

Your knowledge lives as pgvector embeddings in Postgres, not a sealed black box. Technical evaluators get a storage model they can reason about; everyone else just gets accurate answers.

Reuse files already in your workspace

The policy PDF your team uploaded last month is already there. Point an agent at existing workspace files and skip the busywork of maintaining a second copy of every document.

Retrieval that ranks by meaning

Semantic search finds the right chunk even when a customer's wording is nothing like your document's. The agent pulls the closest matches, then writes from them — so answers stay on-topic and on-brand.

Works across your channels

One feature, every platform you're on

New channels arrive as adapters, not rewrites — so this works the same wherever your customers message you.

TelegramLiveFacebookLiveInstagramLiveWhatsAppIn rolloutTikTokComing soonX (Twitter)Coming soon

Keep exploring

Pairs well with

Custom AI Agents

Agents that actually know your business.

Train agents on your docs to auto-answer or draft replies, with a tone you control.

Learn more

Unified Inbox

Every channel in one calm feed.

Assign, snooze, tag, and resolve conversations from every channel without switching tabs.

Learn more

No-code Automation

Triggers, conditions, actions — drag and done.

Build comment-to-DM funnels and auto-replies on a visual canvas, from prebuilt templates.

Learn more

All features

FAQ

Common questions

Ready to run every conversation from one calm inbox?

Start your free trial. No credit card. Connect your first channel in under a minute.

Start free trial Talk to sales