# Parlant

> Open-source conversational AI engine that keeps LLM agents business-aligned and compliant through alignment modeling and granular guidelines.

Parlant is an open-source Alignment Engine for LLM agents that enables developers to precisely control how conversational AI interacts with users in different scenarios. Built by a team of software engineers from Microsoft, EverC, Check Point, and Dynamic Yield, along with NLP researchers from the Weizmann Institute of Science, Parlant addresses the critical challenge of getting LLM agents to behave consistently according to business rules and compliance requirements.

- **Alignment Modeling** provides a structured approach to controlling agent behavior through principles, actions, objectives, and ground-truths that orient agents to specific domains or use cases, replacing rigid flow engines and unreliable free-form prompting.

- **Granular Guidelines** allow developers to define individual clarifications that nudge AI agents on how to approach particular situations, with automatic filtering and selection of the most relevant guidelines for any given interaction.

- **Adaptive Behavior** breaks free from flows and state machines by teaching agents principles to work with, dynamically adapting to natural user interaction patterns while conforming to business rules.

- **Explainable Results** provide extensive feedback on agent decisions and interpretations at every turn, enabling quick troubleshooting and understanding of why agents behave in specific ways.

- **Iterative Development** supports building and shaping agents through conversation and refinement of responses, allowing quick tune-ups whenever unexpected behavior is encountered.

- **Python SDK and React Widget** offer straightforward installation via pip and npm, with a developer-friendly API for creating agents, guidelines, journeys, and tools.

To get started, install the Python SDK with `pip install parlant` and the React chat widget with `npm install parlant-chat-react`. Create an agent with a name and description, then iteratively add guidelines to shape behavior. Parlant is used in regulated financial services, legal assistance, brand-sensitive customer service, healthcare communications, government services, and personal advocacy applications. Licensed under Apache 2.0, the framework is fully open-source with no freemium restrictions.

## Features
- Alignment modeling for LLM agents
- Granular guidelines system
- Adaptive behavior without rigid flows
- Explainable AI decisions
- Iterative agent development
- Automatic guideline filtering and selection
- Enforcement and compliance tracking
- Python SDK
- React chat widget
- Context variables
- Glossaries
- API usage instructions
- Journeys and tools
- Human-readable rationale for decisions

## Integrations
React, Python

## Platforms
API, DEVELOPER_SDK

## Pricing
Open Source

## Version
3.1

## Links
- Website: https://www.parlant.io/
- Documentation: https://www.parlant.io/docs/quickstart/installation
- Repository: https://github.com/emcie-co/parlant
- EveryDev.ai: https://www.everydev.ai/tools/parlant