# Parlant

An interaction control harness for building customer-facing AI agents whose conversations must stay consistent, on-brand and compliant, by narrowing the model's context in real time to what is relevant to the current turn.

Parlant is a harness for building customer-facing AI agents for business-to-consumer and sensitive business-to-business interactions that need to be consistent, compliant, on-brand and traceable.

Instead of overloading a single system prompt or relying on a routed graph, Parlant lets developers define rules, knowledge and tools once, then narrows that context in real time to what is relevant to each turn of a conversation. It focuses specifically on conversational governance and behavioral consistency, rather than on general workflow automation or low-level prompt optimization.

Its design goals are precise control over conversation behavior, such as tone, timing, edge cases, policy constraints and brand voice, and structural prevention of misaligned or unwanted agent behavior. It is written in Python and released under the Apache 2.0 license.

Website: https://www.parlant.io