Import enterprise internal documents, manuals, and regulations into the knowledge base, and Agent can answer questions based on this content — with source citations for every conclusion.
Supports PDF, Word, Excel, PPT, Markdown, HTML, CSV, and more formats. Upload files or enter URLs to import. Three document chunking strategies (fixed-length, recursive, semantic) to match different document structures. Create Markdown documents directly within the knowledge base.
Not simple keyword matching. The system simultaneously retrieves from both 'semantic understanding' and 'keyword matching' dimensions, then uses ranking algorithms to find the most relevant content: semantic vector retrieval captures similar meaning with different wording, keyword retrieval ensures exact hits on proper nouns and IDs. Dual-path results are merged, ranked, and refined through a re-ranking model.
Every answer from Agent is annotated with sources: Citation markers [1][2] pointing to specific document paragraphs Citation alignment score: evaluates how well the answer matches evidence Conflict detection: explicitly flags when multiple documents have contradicting views Confidence assessment: provides answer confidence scores based on evidence quality
Complete lifecycle management: create, edit, and delete knowledge bases. Documents are processed asynchronously after upload, with failed documents retryable. View, edit, and delete individual document chunks with in-chunk text search. Knowledge bases can be bound to specific Agents, auto-retrieving relevant content at runtime.
