# How we launched the Chota knowledge base and where its limits are Canonical HTML: [https://agents.chota.by/en/blog/knowledge-base-launch]() Language: en Product Development Published: 2026-05-12; modified: 2026-07-22. Author: [Chota Team]() — Product editorial team This article was prepared by the Chota team. Product claims were checked against the platform's current capabilities, while external facts were checked against the primary sources linked on the page. Material changes are reflected in the update date. Historical post: capabilities and terms reflect the publication date. The current voice pilot is connected with the team. [Current product]() [Current pricing]() We added a knowledge base so Chota can answer from a company's working materials. Below are the design, value, and limits of the solution. ![Diagram showing documents, CRM data, and a client conversation converging in the Chota AI agent knowledge workflow]() ## Why we moved business facts out of the prompt In early configurations we saw price lists, booking rules, service descriptions, and exceptions placed inside one long instruction. That text is difficult to maintain, and one outdated line can quickly become a wrong answer to a customer. We therefore separated the agent's behavioural rules from business facts. The instruction defines the role and boundaries, while the knowledge base contains documents the team can update independently. - Useful inputs include current price lists, FAQs, regulations, service descriptions, and customer-service rules. - Duplicates, old versions, and conflicting conditions should be removed before upload. - Sensitive or uncertain topics still need a defined handoff to a person. ## How the data flow works During a conversation, the agent matches the customer's question against available context: knowledge-base materials, connected workflow data, and the current conversation history. It then responds or hands the request over when it lacks a reliable basis. We do not claim that uploading any PDF automatically creates a faultless consultant. Output quality depends on document structure, fact freshness, configured boundaries, and realistic test questions before the agent is published. ## Our practical launch checklist We start with a small document set and real customer questions. Materials are added gradually after the answers are checked, making it easier to find the source of an error and avoid mixing multiple versions of one rule. - Assign a person responsible for each document's accuracy. - Test common, ambiguous, and conflicting questions. - Define topics the agent must not answer on its own. - Repeat tests whenever pricing, schedules, or policies change. ## Sources - [Chota product]() [How we select and route models in Chota]() [Chatbot, Telegram bot, or AI agent: how we choose for a business]()