Bookbot, a Czech retailer of second-hand books, utilizes Manticore Search to manage its dynamic inventory. The system handles an index of approximately 5.3 million documents, including 1.29 million physical items currently in stock. By using streaming updates via Kafka, the company ensures that search results reflect real-time availability, pricing, and condition of specific book copies.
The platform features a 'Watchdog' tool that notifies users when specific books become available. This function operates using percolate tables in Manticore, which store 3.5 million user-defined search queries. When a new book is added to the inventory, the system matches it against these queries. According to internal data, this feature accounts for 10% of the company's total revenue.
To improve search accuracy, Bookbot maintains a separate dictionary of 4.52 million entries to correct typos and handle variations in author names or titles. Additionally, the company uses Manticore for a bibliographic database containing 203.44 million records, as well as an internal index of 24.7 million records for administrative tasks and a database of 197,000 pickup locations.
The infrastructure runs on Kubernetes, with separate servers allocated for customer search, internal operations, Watchdog, and bibliographic metadata. This setup allows for independent resource scaling. As of September 5, 2026, the system achieved an average successful query execution time of 7.7 milliseconds for item selection, with 99.9% of these requests processed within 50 milliseconds.
