Give your AI a library, not a guess
Upload the policies, guides, rates and FAQs you already have, connect a read-only data source, and let a Knowledge Agent answer customers from your own material. Replies cite what they read, and live lookups are locked to the person asking.
Answers with references ✦ Live data, customer-scoped ✦ Up to 50 MB per file
Data sources
The most guest-obsessed brands trust Lodgestory to power 50M+ conversations a year
BBingo Hotels
See exactly what the agent reads
An AI that answers from general training sounds confident and is wrong about your business. A knowledge base fixes that, and the inspector proves it: open any document to see the text Lodgestory extracted, the passages it was split into for search with their indexing status, and the original file to download. If a scanned PDF came through garbled, you see it here and re-upload a cleaner copy.
- Instant or background. Text, Markdown, HTML, CSV, JSON and Q&A lists index in a second; PDFs, Word, PowerPoint and scans process in the background, reload-safe.
- Remove any time. Delete a document and the agent stops using it immediately.
- Re-index without downtime. A full rebuild runs in the background while the agent keeps answering.
Parsed text
Passages 0
Built for answers that have to be right
Upload what you already have
Policy PDFs, house guides, FAQ lists, rate cards, a question-and-answer sheet. Drop them in once; the agent searches them on every channel.
Learn moreAnswer from the passage, not the vibe
The agent retrieves the passages that matter, reads them and replies in your wording. If the material does not cover it, a well-instructed agent says so.
Learn morePull live facts, safely
A reference source for rates and availability the agent can query freely; a customer source where every lookup is locked to the verified person in the chat.
Learn moreKeep separate libraries
A Policies base, a Property guides base, a Rates base. Point each Knowledge Agent at the ones it needs and nothing else.
Learn moreSearch, read, then answer, with the reference
When a customer asks, the Knowledge Agent searches your indexed passages for the relevant ones, reads them and, if a data source is connected, fetches live facts too. It can take several steps before replying, and it replies using the material it retrieved. "What is your cancellation window?" comes back with the exact terms from the uploaded policy, not a plausible-sounding guess.
- Meaning-based search. Documents are split into passages and indexed for meaning, so a long manual yields just the part that matters.
- Bounded reasoning. Max steps and reasoning effort are settings on the step, so depth never runs away.
- Any channel. One base powers agents on WhatsApp, email, Instagram and web chat.
What the agent did
Tonight's rate from the table, their booking from theirs
Connect a read-only database. A reference source holds non-personal data such as rates, catalogues and availability, and the agent can run its own read-only queries against it. A customer source holds personal records, and the agent can only run the approved lookups you wrote, each automatically filled with the verified identity of the person in the chat. One guest can never see another's booking.
- Approved lookups only. "My current booking", written once by an admin, picked by the agent, never improvised.
- Withheld where identity is unverified. Inbound email keeps reference lookups and loses customer lookups.
- Both in one reply. The weekend rate from live pricing, the cancellation terms from the policy.
One base per subject, one agent per job
Keep policies, property guides and rates in separate knowledge bases and tick only the ones a given Knowledge Agent step needs. A pre-arrival concierge reads the guides; a reservations agent reads policies and rates; nobody reads the staff handbook. Each base lists its documents and connected sources, and you can have as many as you need.
- Admins curate. Account owners and admins create bases, upload documents and connect sources; agents consume them at runtime.
- The only place it is used is the AI Agent step in Knowledge Agent mode, so what an agent can read is always explicit.
- Part of your plan. AI usage counts against your plan allowance; no separate product to buy.
AI Agent stepKnowledge AgentKnowledge bases this agent may search
Drop the file in; the index takes care of itself
Text-like files are indexed instantly. Scanned or image-based PDFs run through text recognition page by page in the background, and the row says Processing until it is done, reload-safe, so you can navigate away and come back. Individual uploads and deletes update the index automatically; re-index is only for a full rebuild after a large change, and it never interrupts the agent.
- Ten formats. Text, Markdown, HTML, CSV, JSON and Q&A lists instantly; PDF, Word, PowerPoint and scans in the background.
- 50 MB per file, about 2 MB for inline text; larger text files process in the background.
- Failed rows explain themselves. A corrupt or password-protected PDF says so; dismiss and re-upload a clean copy.
Go deeper
Where the knowledge goes to work
Kaylee, the AI agent
Personas per channel, a shared tools library and human approval for the decisions that need a person.
Learn moreBot journeys
The AI Agent step in Knowledge Agent mode is where a knowledge base answers customers.
Learn moreCustom tools
Look up a booking, check availability or open a ticket through your own endpoints, from the same agent.
Learn moreIn practice
The policy concierge
A villa group uploads its cancellation policy, check-in guide and the Wi-Fi and pool FAQ. A WhatsApp Knowledge Agent now answers "what is your cancellation window?" with the exact terms from the uploaded policy. A hotel connects its bookings database as a customer source with one approved lookup, and "what time is my check-in?" is answered from the guest's own booking, scoped to their verified number, without exposing anyone else's record.
Governed by design
Customer-scoped lookups are locked to approved queries and bound to the verified identity of the person in the chat. The agent cannot run arbitrary queries against personal records and cannot look up someone else, and those lookups are withheld on channels where identity cannot be verified.
Visit the trust centerGuides for the people who curate the library
Create, upload, inspect, connect a source, point a Knowledge Agent at it, test.
Every setting on the step that consumes your knowledge bases.
Per-channel voices, system prompts and function-calling tools.
Ready to answer from your own material?
Upload one policy, ask a question in the test chat, and watch the answer come back in your wording.
Questions, answered
The things teams ask before they trust an AI with their policies.
Text, Markdown, HTML, CSV, JSON and question-and-answer lists index instantly. PDFs, Word, PowerPoint and scanned documents are read and indexed in the background, with text recognition for scans. Up to 50 MB per file.
Open the document inspector. The parsed-text pane shows exactly what Lodgestory extracted, the passages pane shows the pieces it was split into for search with their indexing status, and the original stays downloadable. If a scan came through garbled, you will see it and can re-upload a cleaner copy.
No. A customer data source only runs the approved lookups you define, and every lookup is automatically scoped to the verified identity of the person in the chat. The agent cannot write its own query against personal records or look up anyone else, and these lookups are withheld entirely on channels where identity cannot be verified, such as inbound email.
No. A plain conversational AI agent works without one. Add a knowledge base when answers must be grounded in your own documents or live data: cancellation windows, check-in times, tonight's rate, a guest's own booking.
In bot journeys, through the AI Agent step set to Knowledge Agent mode. Tick one or more knowledge bases, write the persona, choose the model and deploy. The same base can power agents on WhatsApp, email, Instagram and web chat.
Individual uploads and deletes update the index automatically. Re-index is only for a full rebuild after a large change, and it runs in the background with no downtime: the agent keeps answering from the current index until the new one is ready.