Qlik says its Qlik Answers platform uses Amazon Bedrock and a multi-layer agent architecture to deliver sourced responses across enterprise data.
Qlik has outlined the architecture behind Qlik Answers, its AI assistant for searching enterprise information, which the company says serves more than 40,000 customers. The product became generally available in February 2026. Since then, Qlik reports that its Discovery Agent has surfaced more than 100,000 discoveries, and that a majority of Qlik Cloud accounts with agentic tools enabled are actively using them.
The system routes questions according to the kind of information needed. Answers can draw on structured analytics, documents, knowledge bases, glossary entries or automations. Qlik separates request routing from response generation, then uses specialist agents and a dedicated analytics path for questions involving structured data. For document search, it uses Amazon OpenSearch Service to index and retrieve content.
Amazon Bedrock provides access to models for chat, streaming, embeddings and reranking through Qlik’s own language-model gateway. Qlik says this lets its product teams choose models for particular tasks without tying application logic to a single model. Amazon Bedrock Guardrails are applied to requests and responses, including checks for prompt injection, personally identifiable information, secrets and disallowed topics. The system also checks generated answers against source material; responses to unstructured-data questions can include citations to retrieved content.
Qlik says the design also addresses regional data requirements. Bedrock cross-Region inference supports deployments across 11 Regions, while the company can host a model on Amazon SageMaker AI when it is not yet available in the needed Region through Bedrock. Qlik forecasts model capacity three to six months ahead of major launches and compares projections with usage after rollout.
Customer examples in the post include Lintech International, which indexed more than 17,000 technical documents and reported 75 percent faster response times, returning as much as seven hours a week to business managers. Bystronic reported deploying a chatbot in 15 minutes for questions about operational data. Qlik says it is evaluating Amazon Bedrock AgentCore for selected workloads, while continuing to develop its own orchestration framework and model-evaluation process.