Oliva utilizes theLangGraph Graphical Workflowdesign, its multi-intelligent body system contains three key components:
1. Division of roles
- supervisory smartphone: Acts as a central scheduler, evaluating task types and routing them to specialized intelligences
- Search for Intelligentsia: Responsible for hybrid semantic/vector search of Qdrant database
- Generating Intelligentsia: Formatting of processing results and natural language responses
2. Collaborative processes
- User voice commands converted to text by Deepgram
- Supervisor parses the intent and triggers the corresponding workflow node
- Search Intelligence calls LangChain+Superlinked for multi-dimensional retrieval
- Generating intelligences to transform raw data into user-friendly outputs
3. Extension mechanisms
Developers can:
- exist
app/agents/implementations/Adding New Intelligence to - pass (a bill or inspection etc)
langchain/tools/Define specialized tools - modifications
langgraph/config/The process configuration file in the
This architecture is particularly suitable for scenarios that require complex task decomposition, where each intelligence focuses only on a specific capability domain.
This answer comes from the articleOliva: a voice-controlled multi-intelligence product search assistantThe
May not be reproduced without permission:AI productivity tools " How does Oliva's multi-intelligent body architecture work?
