FAQ

Parrotbox.AI lets you enter a key code for your organization’s preferred AI account (e.g. ChatGPT, Gemini, Claude), which will be used to process Parrotbox interactions in the background. Most AI platforms offer professional-tier accounts where your data will not be used to “train” models, which means conversations will be secure and private. You can even choose to save sessions-in-progress to an outside database controlled by your organization, so not even Parrotbox’s team can see them.

Parrotbox.AI makes it easy to deploy AI interactions everywhere. You can either provide users with a direct link to an AI interaction or embed interactions (as an iframe) within websites, apps, and e-learning modules with a responsive interface.

You can also download a “shell” e-learning module containing a Parrotbox.AI module that can be uploaded to a learning management system (LMS) and report scores via SCORM.

User management can be handled either via Parrotbox’s platform or by your LMS / another platform via API integration.

The short answer is “yes.” You can upload documents to a “Retrieval Augmented Generation” (RAG) provider, which lets the AI search documents on an as-needed basis. Many AI platforms have built-in RAG functionality, or you can use an outside service like Vectara.

For other data sources, Parrotbox offers an API which your developers can use to pass data from an outside database or software application into Parrotbox, where it becomes accessible to the AI (e.g. you could feed actual customer records into a sales or customer service training simulation).

When designing interactions for clients, our quality assurance (QA) standard is to achieve 32 to 64 consecutive sessions with generally acceptable performance and no “unacceptable” behavior by the AI.

That said, AI prompts are fundamentally different from traditional software programs – AI inherently takes creative liberty with any instructions you provide it (which is part of its unique strength), so there is never a 100% guarantee that AI models will respond to instructions the same way every time. Targets for consistency are usually set on a case-by-case basis, with a typical standard being 95% or higher rate of acceptable behavior.

One of Parrotbox’s strengths is the ability to use multiple AI models within a single interaction, letting each model handle the specific tasks for which it’s best suited.

Currently we support all ChatGPT models up to 4o. Support for Claude, Gemini, ChatGPT o1, Perplexity, LLaMa and other platforms is on our near-term roadmap, prioritized by customer demand (that said, ChatGPT currently outperforms most other models for training and coaching tasks by a significant margin).

If you sit down and type “create a sales training role play” into a standard AI chatbot, you might or might not get an acceptable experience. However, if you want to deliver a consistent, high-quality experience to a large audience, then tools like Parrotbox.AI offer many advantages. With Parrotbox, you can create more complex prompts (with tens of thousands of words of instructions), set specific ending conditions and time limits on sessions, capture session data (including SCORM), manage usage, and use special commands and features that provide more control over the AI’s behavior and the end user experience.