Skip to content

Feature · Knowledge base

A knowledge base chatbot your users can talk to.

Upload the documents your team trusts, and a PyAI voice agent searches them on each turn to answer aloud.

Get a sandbox key$50 credit. No credit card required.
90% grounded answers in PyAI testsUploads up to 25 MB per fileMeaning plus keyword searchOmni $0.05/min, no platform fee

What PyAI gives you for knowledge base

  • A knowledge base chatbot on PyAI takes each source as a file up to 25 MB, a single URL, a site crawl with a page limit, or pasted text.
  • Retrieval blends meaning-based search with keyword search, fused by rank, so an exact product code like C-12 and a loose paraphrase both find the right passage.
  • A knowledge base chatbot also needs a fallback, and the agent's instructions decide whether it says the answer is not covered, takes a message, or transfers to a person.

How does the agent answer, and what if the answer is missing?

Three invented sessions with a holiday rental company's booking app show retrieval in action. The lines are an example script, not a recording.

Pick a moment, then press play

UserAI voice agentPyAI retrievalYour team
Ready00:00

How does an AI-powered knowledge base reach the user?

PyAI voice infrastructure keeps your files, the search and the live conversation in one system. The agent looks up the answer while the user waits, with no other vendor in between.

Add sources

Create a knowledge base, then add files, URLs, crawls with a page limit, or pasted text through the Console or the API.

Index

A worker fetches each source safely, extracts text, splits it into overlapping chunks, embeds them and marks the document Ready or Needs attention.

Bind

Attach one or more knowledge bases to the agent with weights in the Console or through the API, or let the org default apply.

Retrieve per turn

On each user turn the agent pulls the top passages with combined meaning and keyword search, then speaks an answer built from them.

Passage found

The agent answers in a sentence or two, using the wording of your document for prices, dates and limits.

Nothing relevant

The agent follows its instructions for gaps: admit it, collect details, or hand off to a person with the hand-off tool.

Where does a voice knowledge base pay off?

A voice knowledge base works well for questions where the answer is already written down. Here are five examples across apps, websites, kiosks and one phone line. Each one uses only features PyAI has today.

Holiday rental app

Policies and amenities

A rental company indexes its cancellation policy PDF and an amenities spreadsheet per cabin. Guests asking the booking app about deposits or dishwashers get answers from those two sources without reaching reservations.

In-app software support

Help center crawl

A SaaS vendor crawls its help center with a page limit, so the voice assistant inside the product answers login and invoice questions. Outage reports, which the help center does not cover, go to the engineer on duty.

Council phone line

Bin collection and permits

A council office pastes collection schedules and permit rules as text. Residents calling about missed bins hear the exact rule, while complaints are taken as messages.

Clinic intake tablet

Preparation instructions

A clinic uploads fasting and preparation sheets for each procedure. The waiting room tablet reads back the instructions that match the procedure the patient names and never offers medical advice beyond them.

Store website bot

Returns and sizing

An online shoe store crawls its returns policy and size guide pages. Shoppers using the site voice widget hear the 30-day window and fit advice the store has published, and damaged parcel claims are passed to the customer service team.

Knowledge base chatbot questions

How does a knowledge base chatbot avoid making up answers?

A knowledge base chatbot stays accurate by answering only from retrieved passages and saying so when nothing relevant is found. A common complaint is bots that guess instead. A PyAI agent's instructions can require it to admit gaps and transfer or take a message, and PyAI tests found 90% of answers grounded.

How do you keep a voice agent's knowledge base current?

Keep a voice agent's knowledge base current by updating the source documents whenever a policy, price or product changes. On PyAI, recrawl changed pages or re-add updated files through the Console or the documents API, then check that each one shows Ready. Teams with a docs pipeline usually trigger that update automatically on publish.

How is a knowledge base AI different from a general chatbot?

A general chatbot answers from its training data, so it can sound sure about policies it has never seen. A knowledge base AI retrieves passages from approved company content first and grounds the reply in them. In PyAI tests, 90% of voice agent answers were grounded in content bound to the agent.

Can the PyAI knowledge base chatbot answer typed chat?

No typed chat widget appears in PyAI documentation. PyAI agents answer by voice through the website voice widget, your own app or a phone line, and visitors open the widget with a talk button. Teams that also want typed chat on their site need a separate chat product, which can draw from the same source documents.

PyAI is voice infrastructure. It hands you the audio and transcript, and does not replace your own models or review tools.

Your documents already hold the answers. Let every user hear them, read aloud by an agent.

Start building with $50 creditSandbox keys never bill, and you can start without a credit card.

One stack: Omni, Hear, Speak, Recap and Trace.