Conversation Engineering

Parlant 3.3 introduces tag-based relationships, numeric guideline priorities, transient guidelines, Agent.utter(), and easier metadata management from tools.

Parlant 3.2 introduces streaming message output, entity labels with session propagation, and improvements across guidelines, journeys, and canned responses.

Parlant 3.1 introduces Emcie for cost-optimized inference, and substantial improvements across guidelines, tools, canned responses, and observability.

Not all instructions carry the same weight. Learn how criticality levels let you control the accuracy-cost tradeoff in conversational AI agents.

When we started building guideline matching, we thought semantic similarity would be enough. We were wrong. Here's what it actually takes to make AI agents reliably follow hundreds of business rules.

LangGraph's router and multi-agent patterns are powerful for orchestration, but they create inherent challenges for free-form conversational AI. Here's why, and how Parlant solves it differently.

Many people are asking if Parlant is like DSPy. Let's clear up the confusion and explain what each framework actually does.

Introducing Parlant 3.0 with massive performance improvements, enhanced journeys, canned responses, and comprehensive production features for enterprise deployment.

A deep dive into the challenges and solutions of designing APIs for agentic applications.

Explore the innovative techniques and strategies we've implemented to enhance the accuracy of embedding-based retrieval systems.

Discover how our team's journey from building mission-critical systems led to the creation of Parlant, an open-source Conversational AI engine designed for reliable customer interactions.

In 2025, the most effective way to build customer-facing AI agents combines the adaptability of LLMs with the structure and precision of traditional conversation design.

The challenges of LLM hallucinations in large-scale conversational AI and strategies for mitigating them.

Conversational AI has evolved dramatically over decades. From rule-based chatbots like ELIZA in the 1960s, through hybrid ML+rule frameworks, and, recently, LLMs. Now, state-of-the-art Conversation Modeling engines marry the generative power of LLMs with structured guidelines, allowing richer interactions that remain controllable, easier to develop iteratively, and more scalable to test in real-world scenarios.

In the market for generic RAG frameworks, the different providers are fighting over who can provide 67% accuracy versus 65%. And when you run an off-the-shelf RAG framework on your use case, it will end up closer to 50% accuracy. Is this the best that the industry can do?

A commentary upon Yann LeCun's key note at AI Action Summit, along with some supplementary explanations on how LLMs work under the hood

A peek under the hood into how LLMs work

A deep dive into today's prevalent methodologies, the challenges that come with each of them, and where the future may lie.