Parlant

Parlant is an open-source interaction control harness for building customer-facing AI agents that behave consistently, predictably and on-brand. Instead of stuffing ever more instructions into a system prompt, Parlant’s engine matches only the guidelines, knowledge and tools relevant to each conversational turn. Positioned as an open-source alternative to Ada, Decagon and Sierra, it has earned about 18,300 stars on GitHub.

Overview

You define agent behavior in code: guidelines as condition-action rules, journeys as multi-turn standard operating procedures, canned responses for critical moments, a glossary of domain terms, and tools that fire only when their observation matches. Relationships between guidelines, dependencies and exclusions, keep the assembled context narrow, so adding hundreds of rules makes the agent more capable without the usual prompt-overload collapse. Every decision is traceable through OpenTelemetry logging, which matters in regulated industries; Parlant reports production deployments in banking and other stringent organizations. It is LLM-agnostic, ships an official React chat widget, and installs from PyPI.

Key Features

• Guideline engine that matches behavioral rules to each turn instead of one giant prompt
• Journeys for multi-turn procedures that adapt to how the customer actually interacts
• Canned responses that eliminate hallucination at compliance-critical moments
• Contextual tool calling, triggered only when the matching observation holds
• Full explainability with OpenTelemetry traces of every guideline match and decision
• LLM-agnostic Python framework with an official React chat widget

💳 Pricing

Parlant is completely free and open source under the Apache-2.0 license. You deploy the framework yourself and pay only for the LLM APIs your agents call.

Who Is It For

Parlant is for product and engineering teams building customer support, sales or advisory agents where tone, accuracy and compliance matter too much to leave to a raw system prompt.