Oliva is a multi-agent assistant that helps users find products on Qdrant database using Langchain and superlinked.

1. Install project dependencies:

uv sync

This will create a virtual environment in .venv and install all required dependencies.

2. Livekit account

Create a Livekit account in Livekit Cloud and get LIVEKIT_URL, LIVEKIT_API_KEY and LIVEKIT_API_SECRET.

LIVEKIT_URL=wss://your-project.livekit.cloud
LIVEKIT_API_KEY=secret
LIVEKIT_API_SECRET=********

3. Environment variables

Before running any Python scripts, set the following environment variables:

cp .env.example .env

4. Qdrant

Use docker to run Qdrant, set an API key wherever you want:

docker run -p 6333:6333 -p 6334:6334 \
    -e QDRANT__SERVICE__API_KEY=******** \
    -v "$(pwd)/qdrant_storage:/qdrant/storage:z" \
    qdrant/qdrant

make oliva-start

Use Agent playground and connect with your Livekit project to interact with the voice assistant. If you prefer run locally, download the repo Agent playground and run npm run start.

oliva/
├── app/
│    ├── agents/
│    │   ├── implementations/     # Individual agent implementations
│    │   ├── core/       # Base classes and interfaces for agent components
│    │   └── langchain/
│    │       ├── base/      # Base LangChain integration classes
│    │       ├── config/    # LangChain configuration
│    │       ├── edges/     # Edge conditions for workflow routing
│    │       ├── nodes/     # Node implementations (agent, rewrite, generate)
│    │       └── tools/     # LangChain-specific tools
│    ├── voice_assistant/
│    └── utils/              # Shared utilities

The project follows a modular architecture implementing an agentic RAG (Retrieval-Augmented Generation) system:

1. Agent Components (app/agents/)

   • agents/: Contains specific agent implementations
   • core/: Defines core interfaces and abstract classes for:
     • State management
     • Node implementations
     • Edge conditions
     • Tool interfaces
     • Graph workflow definitions

2. LangChain Integration (app/agents/integrations/langchain/)

   • Provides LangChain-specific implementations for:
     • Document retrieval
     • Tool operations
     • State management
     • Workflow nodes and edges

3. Voice Assistant (app/voice_assistant/)

   • LiveKit integration
     • Voice interface implementation
     • Speech-to-text and text-to-speech capabilities

4. Utilities (app/utils/)

   • Shared helper functions
   • Common utilities used across modules

The system implements a graph-based workflow where each agent processes state through a series of nodes (functions) connected by conditional edges, supporting dynamic routing based on the agent's decisions.

Langchain workflow by superlinked

make agent-search-by-superlinked

Langchain workflow by json file

make agent-search-by-json