Skip to main content

A more sensible feedback loop

Parlant gives you structured control over your agents' behavior through a system of reliably-enforced small, focused guidelines. When you notice unwanted behavior, you can add precise refinements to address it, while Parlant automatically checks for conflicts in your design. Because great interactions evolve through experience, Parlant gives you a rapid feedback loop that lets you iterate safely, without risking existing functionality.

Predictable and conformant

It’s hard to let go and trust an LLM-based agent, especially if it tends to get "creative." Parlant helps you stay sane by monitoring and double-checking every agent response to ensure that it consistently adheres to your configurations, resulting in correct and conformant responses. It also filters out inappropriate and malicious user inputs from ever reaching your agents, making it a good fit for sensitive use cases that require compliance.

A guided approach to tool-calling

Reliable integrations need more than just giving agents access to your APIs, because APIs are hardly ever self-explanatory. Parlant’s vendor-agnostic, turbo-charged tool-calls implementation lets you show agents the why, the when, and the how of using your APIs. Better yet, you can give different instructions on how to use the same tool in different scenarios for different goals.

Create an agent
Create a session
Create a guideline
Interact with the agent
import { ParlantClient } from 'parlant-client';
const client = new ParlantClient({ environment: SERVER_ADDRESS });
const agent = await client.agents.create({
name: "Scooby",
description: "Technical support specialist",
});

Parlant's Philosophy

Guidance is Needed

LLMs know countless different ways to approach any situation. We assert that without clear guidance, they will often make arguably reasonable, but nonetheless arbitrary choices. Aligning them with our specific intentions should be quick, easy, and reliable.
Human-Centric Approach

Training AI agents should mirror how we train human agents: through clear feedback loops and systematic improvements. This makes the process intuitive for business stakeholders while maintaining technical rigor.
Design for Control

Software should create alignment and control structures that enable practical development processes and methodologies. We believe it's the hardware that should and will catch up to make such processes fast and economical.