Flowrite utilizes an autoregressive language model GPT-3 to produce human-like text from short samples. The model is developed by an AI research and deployment company OpenAI and is made up of 175 billion parameters, making it the most advanced language model ever created.

Flowrite does not read any data from your personal accounts – the completions are created based solely on your short input, Flowrite's own algorithm, and the GPT-3 language model. The only exception to this is when you are, for example, replying to a message or following up on one. In that scenario, Flowrite uses the previous message as a sample to perceive the context – but only with your explicit permission and awareness.

We are developing our own proprietary technology that learns from your completions and gathers quantitative feedback from you in the extension interface. Flowrite remembers your style and phrasing the next time you use it in a similar context.

We are doing our best work to make Flowriting a safe and unbiased experience for all users. Safety plays a major part in our development roadmap and we will be safeguarding against unwanted behavior when going live with our first users.

With that said, we acknowledge the shortcomings of GPT-3's current version and are actively in contact with OpenAI to stay at the forefront of their safety development. As an example, all GPT-3 -powered projects looking to go live are required to go through a production approval process where the possible risks are assessed.