| PDF parsing & chunk control | Deep built-in document layout recognition with template-based chunking pipelines. | Configurable multi-parser routing (MinerU, Mistral OCR, LlamaParse, MarkItDown), Jev DP semantic chunking, plus manual visual chunk merge/split with incremental re-indexing. | PDF parsing |
|---|
| Knowledge graph governance | Graph-assisted retrieval over extracted document entities. | Tri-index RRF (vector + BM25 + graph) with custom domain entity/relation schemas, auditable & reversible alias merges, and v2 topic discovery preview. | Knowledge graph |
|---|
| Agent execution, SQL & evaluation | Document Q&A and retrieval-centric agent workflows. | Offline Code Mode sandbox, read-only NL2SQL Database KB with 3-way schema diff rollback, execution diagnostics, and immutable versioned evaluation reports. | Agent diagnostics |
|---|
| Protocols, IM & HA operations | API/SDK integration with self-hosted containers. | Built-in Streamable HTTP MCP Server (20 tools), OpenAI-compat API, A2A 1.0, 5 native IM channels, dual-plane RBAC, and Kubernetes + KubeBlocks Enterprise HA. | MCP & protocols |
|---|