Most "chat with your docs" tools stop at a single vector search and a prompt. The harder problem in enterprise settings isn't retrieval — it's making retrieval, branching logic, and external API calls observable and reproducible enough to trust in finance, HR, or legal workflows. This is where the visual flow editor matters more than the RAG itself: every node, every retrieval call, and every model decision is inspectable.
What Sets It Apart
- Hybrid retrieval with reranking and live, auto-cleansed data is wired to reduce hallucination, not just fetch the nearest chunks — the retrieval quality is treated as a first-class config, not a black box.
- The block-based "Flow" editor lets you compose conversational logic, plugins, and RPA nodes visually, so a non-engineer can reshape an agent without touching code.
- Dual-direction MCP support and full call-chain logging mean the agent can both call and expose tools, while every step stays debuggable at the node level — rare for self-hostable RAG platforms.
Who It's For
Great fit if you need governed, on-prem AI agents over proprietary knowledge — SSO/RBAC, self-hosting, and audit-grade call logs are built in, which is why it lands in regulated teams. Look elsewhere if you want a lightweight personal chatbot or a hosted assistant with zero ops: running it well means managing your own deployment, vector store, and model providers, and the visual editor adds a learning curve over a plain prompt.
