Product

A knowledge base for your AI sales agent: the facts it answers from

The knowledge base holds what the agent should know and never guess: prices, policies, opening hours, objection answers and product details. You add documents, URLs and text. The agent searches them by meaning when it replies, and you choose which agents can see which collection. The persona holds who the agent is, and the knowledge base holds the facts.

Published by Sellinger AI. Numbers and settings checked on 19 October 2026 against the fact register. Report an error.

What goes where

In the persona

  • Who the agent writes as and in what tone
  • What you sell and to whom, as positioning
  • The goal and the rules for when to stop

In the knowledge base

  • Prices, plans and discounts
  • Opening hours, policies and delivery terms
  • Links and proof points
  • Answers to objections you have heard

The split is deliberate. A price written into a persona goes stale in every prompt the moment it changes. The same price in the knowledge base is one article to edit.

How it works

  1. 1

    Add

    Documents, URLs or text

    Create a collection, add articles, or quick add a URL when you create it.

  2. 2

    Index

    Articles are split into chunks

    Each chunk is embedded, so the agent finds an answer by meaning and not by exact words.

  3. 3

    Link

    Choose the agents

    A collection is only visible to the agents and personas you link it to. With exactly one active agent, a new collection links to it automatically; otherwise you are told it is not connected yet.

  4. 4

    Keep current

    Rebuilds on change

    Adding, editing, deleting or unpublishing an article marks the collection stale and rebuilds the digest the agents use, so a changed price reaches the next reply.

From Claude or ChatGPT

The connector has a `search_knowledge_base` tool that checks what an agent was told about a topic. It matches your text as a literal substring of titles and bodies, not by meaning, so a miss means that wording is absent and not that the topic is uncovered. Articles are edited in the app. See the connector.

Limits

  • The agent can only answer from what you give it. If a fact is not there, the right behavior is to hand the question to a person, which a needs attention rule can enforce.
  • Public web pages added by URL must be public. Private and internal addresses are refused.
  • Sellinger has no published measure of how much a knowledge base improves answers, so none is claimed.

Common questions

Do I need a knowledge base to start?
No. A persona alone works, but price and policy questions then go unanswered or to a person.
Can I add a PDF?
You can add documents, URLs and text. Check the app for the current file types.
Does the agent quote it word for word?
It answers from the matching chunks in its own words. Put exact wording, such as a legal line, in the rules as well.

Sources and method

frontend/src/lib/knowledge-base (article chunks, staleness, schedule, auto link), the knowledge base page and the connector's search_knowledge_base tool, checked 19 October 2026.

Related: agentic CRM, MCP connector, AI sales agent.

See it on your own accounts

Start with a 30 day free trial, no card. Nothing is sent without your approval unless you turn that on.

Unit4ShellyRocket.ChatGetAcceptPeopleForce

Teams that have used Sellinger

Start free trial