parlant

Overview of parlant

What it is

Parlant is an open-source interaction control harness for building customer-facing AI agents. It focuses on controlled, consistent, and predictable LLM interactions through context engineering. The framework is designed to be production-ready for enterprise-grade B2C and sensitive B2B interactions that require consistency, compliance, brand adherence, and traceability. It serves as an alternative to systems like Ada, Decagon, or Sierra, and is deployed in production at organizations including banks.

Key features

• Defines agent behavior in code (not prompts) with rules, knowledge, and tools.
• Dynamically narrows context per conversational turn to only immediately relevant instructions and tools.
• Supports Observations, Guidelines, Journeys (SOPs), Retrievers, Glossary, and Variables.
• Offers two output modes: Fluid Output (generated message) and Strict Output (canned message).
• Built-in matcher system to automatically apply guidelines based on conditions.
• Exclusion mechanism to prevent conflicting behaviors from entering context.
• Emphasis on maximum control over conversation experience and prevention of unwanted behaviors.

Use cases

• Customer-facing support, sales, onboarding, and advisory agents where tone, accuracy, and compliance are critical.
• Regulated or high-stakes domains such as finance, insurance, healthcare, and telecom requiring explainable and auditable responses.
• Teams with dozens or hundreds of behavioral rules who need their agent to behave reliably in front of real customers.

Getting started

See the repository README for installation and quick-start instructions.

License and pricing

Parlant is licensed under Apache-2.0 and is open source with no stated pricing.

Based on the project's public README and repository metadata.