How emcie-co/parlant works
Project Type and Purpose
• This is a full-stack AI agent framework (“Parlant”) designed to let developers define conversational AI agents via natural-language guidelines.
• The backend is a Python server exposing REST/WebSocket APIs, the core engine applies guidelines, and the frontend is a React chat widget.
• Key use cases: production chatbots in financial, healthcare, e-commerce, legal domains with guaranteed rule compliance.
High-Level File Structure and Architectural Patterns
• Top-level “src/parlant”
• adapters: connectors to databases (JSON file, MongoDB, in-memory), NLP providers (OpenAI, Anthropic, AWS, Azure, HuggingFace, etc.), vector DBs (Chroma).
• api: HTTP/WebSocket endpoints, React chat widget under api/chat.
• core: business logic, guideline matching engine, event emission, NLP helper layers, persistence abstraction, tool integrations.
• bin: CLI entrypoints (client.py, server.py).
• sdk.py: Python client library.
• Tests organized per layer (adapters, api, core, sdk) confirming a modular clean-separation architecture.
• Deployment: Docker in .devcontainer, GitHub Actions CI, build/publish scripts.