Advantages

Out-of-the-box usability

Document Q&A can be made ready immediately after documents are uploaded. You can switch between different LLMs on the model management page.

High accuracy

KooSearch uses a high-accuracy text embedding model, which has a high ranking on C-MTEB (Chinese Massive Text Embedding Benchmark). It also provides built-in reranking and search planning services; and supports multiple search methods, such as keyword-based search, vector search, and hybrid search. Based on deep document understanding, KooSearch can accurately identify the layout of documents from complex unstructured data, including titles, paragraphs, line breaks, headers, and footers.

KooSearch supports flexible text slicing—It can slice documents based on their directory structure or custom rules. The whole slicing process is visualized and can be manually adjusted to ensure end-to-end accuracy.

Figure 1 High ranking on C_MTEB

High Performance

The KooSearch uses a CSS vector database, which has a high ranking on ANN-Benchmarks. This database supports multiple types of indexes, such as Flat, Graph, IVF, IVF_Graph, and PQ, and is fully compatible with the Elasticsearch ecosystem. It achieves a perfect balance between performance and precision.

Security

KooSearch supports physical multi-tenancy for secure tenant isolation; fully managed services; fine-grained permission management, including LDAP integration; and knowledge base isolation.