Introducing Convai

Purnendu Mukherjee
August 20, 2023

Why did we start Convai?

Convai, pronounced as “convey”, was started with the aim of empowering virtual world developers with the latest in Conversational AI. Every time there is a technological shift, we, as humans, change and adapt to it forever. Large Language Models (LLMs), like ChatGPT, are one such application that has already caused fundamental changes in human behavior. People who have used it are never going back to their old ways again. Innovators and early adopters know that Generative AI will have a significant impact on almost every industry. Although Generative AI applications are seeing fast adoption in both consumer and enterprise use cases, the virtual world industry is yet to reap the benefits. Why is that the case?

Powering Non-Player Characters (NPCs) with human-like conversation capabilities in XR and games seems like an obvious use case. However, there are hidden bottlenecks that a developer encounters when attempting to integrate Conversational AI tech with virtual environments. Having experienced the challenges myself as an engineer and from hundreds of conversations with virtual world developers, it’s clear that there is a missing layer of infrastructure that developers need. This layer of infrastructure should not require developers to stray too far away from their existing game development process. Ideally, it should be easy to use and enable developers to create and manage their characters’ dialog capabilities, as well as integrate with their own NPC assets. Additionally, it should take care of latency issues and scaling challenges of real-time generative speech and language models, without worrying about LLM costs that can disrupt existing economics. We developed Convai to address that missing layer.

Why should you, as a developer, consider building on Convai?

  1. Ease and Speed of Development: As developers ourselves, we understand the pain of dealing with difficult interfaces and unclear documentation. While we are just beginning to perfect this process, our commitment is to build the simplest development interface where you can create AI characters and deploy them in your environment/game in minutes.
  2. Innovation Factor: Given our ambitious target of human-like intelligence for NPCs, you can count on us to innovate relentlessly. Here are a couple of innovative features that we have already published, and we’d love you to try out:
    Knowledge Bank: This allows you to provide as much information/knowledge as you’d want for your character. Make them an expert on a topic or provide them with the lore of the world you’re building. While this is in early beta, it helps mitigate the ‘hallucination of information’ issue that plagues the best LLMs.
    Scene-aware actions: Your character inside the world can take the scene metadata as input and, based on conversation/commands, can generate the right set of actions to carry out complex tasks and strategies.
    These features will continue to improve with time, and you will find more innovative features going forward.
  3. Optimized for Scale and Latency: We are building for a future where games and virtual world experiences are powered by AI as a core component. To achieve this, we are designing our APIs and plugins to not only enable real-time natural responses (low latency) but also to handle millions of interactions that can happen within your application or game. You can count on our infrastructure to scale with you as your app or game gains popularity.
  4. AI Moderation Control and Balance: When it comes to AI moderation, what should be flagged and filtered by a moderation layer is very context-dependent. For example, an AI tutor for kids would typically require quite strict moderation for its responses, regardless of the input. In this case, all toxicity filters are set to maximum, with potential violations being flagged, and the AI character disengaging from the conversation when flagged. Games, on the other hand, may require a more nuanced approach, where careful moderation and thoughtful use of moderation classifiers should be implemented. For example, interactions in a game with a villainous character with nefarious objectives could generate responses that would likely trigger some moderation flags. Therefore, rating systems used for movies and current games (e.g. PG13, R, etc.) may be useful. We recognize that sensitivity to these issues is required when attempting to balance the needs of artists and storytellers and those of their audiences. In general, unless we feel very confident about our moderation classifiers, we prefer to employ stricter moderation rules, taking the ‘better safe than sorry’ approach.
  5. Cost: Our goal is to provide you with the highest value in the market while being the best in its class. We offer a free tier for you to get started with development and thereafter we offer transparent pricing plans that enable you to start small and grow as your budget allows. If you simply take the market rate of underlying individual services such as Speech-to-Text, ChatGPT, and Text-to-Speech, etc. you will find that our per-interaction cost is actually cheaper than the combination of those. The rest of the platform, infrastructure, and support effectively is free. Apart from our subscription tiers, we offer various other options via Enterprise and partnership plans that reduce the time to market for your product by an order of magnitude. Our mission is to enable you to create delightful experiences for millions of users and not let cost be the barrier to that.
  6. Support: Although we might work on virtual worlds, most of us are still firmly grounded here on Earth. Our goal is to make Convai the most developer-centric company. We offer comprehensive support for developers, providing documentation, sample code, and tutorials to help you get started. Additionally, Convai’s support team is available 24/7 to assist you with any issues that might come up.

What do we believe regarding the role of conversation AI in games and virtual worlds?

We believe that AI will increase the utility of virtual world applications by orders of magnitude, where learning becomes more engaging, and AI characters become an integral part of the experience. Games will have bots that can play alongside human players while interacting, strategizing, and executing their commands. NPCs will interact in an open-ended manner while staying on course for a narrative they were meant to be a part of.

In a world where AI-driven conversations enable our applications and games with a whole new world of possibilities, the experience becomes a lot more fun and engaging.

If you share the same belief, sign up at, and let’s start building. If the work at Convai excites you, please apply as we are hiring!