Agent Canvas
Every knowledge base has a visual agent canvas (built on React Flow) for orchestrating the conversation pipeline.
Node Types
| Node | Color | Purpose |
|---|---|---|
| Intent Classifier | Violet | Route user input to different branches |
| Retrieval | Emerald | Retrieve relevant content from the KB |
| Condition | Amber | Branch based on conditions |
| LLM | Blue | Call the language model to generate a reply |
| Direct Reply | Cyan | Return fixed text |
Templates
- Minimal template: LLM + direct reply — for pure-chat scenarios (no documents or embedding model needed)
- Standard template: retrieval + LLM — for knowledge-base Q&A
Configuration
- System prompt (LLM node)
- Opening message (app level)
- Temperature, top_k, rerank toggle
- LLM provider selection
Run Test
The test drawer on the right lets you chat in real time to verify the pipeline, and shows today's message quota usage.
During execution, nodes and edges highlight in real time as the pipeline progresses; closing the drawer restores the canvas to its original state.