Upstream let's users build and deploy AI chatbots trained with their own data. User-trained chatbots are a powerful way to build specialized AI assistants. This means you can combine multiple data sources and train an AI to answer questions specific to those sources.

It's super easy. Here's how:

• Upload files from your computer.
• Select them to train.
• Soon after, your chatbot will be available for use.

Here's a few examples. We're excited to see other creative ways you might use them too.

• Business chatbot. Businesses train a chatbot on their company's website to add a new way for their customer's to learn about their products and services.
• R&D chatbot. Teams aggregate research documents and papers into a chatbot to compare and discuss approaches and ideas which helps them move faster.
• Translation chatbot. Teams are able to train a chatbot on their technical documentation to help non-technical people understand their products.
• Hobbyist chatbot. Hobbyists train a chatbot on their hobby's content to showcase it.

We use whichever is the most powerful available AI model. Today that is OpenAI's GPT-3.5-turbo, which is the same as ChatGPT.

We currently support PDFs and HTML although soon we will add general text files and audio.

Language models when conditioned with content are more likely to say factual things. That said, AI can make mistakes and you should use your own judgement before making decisions or taking action based on Upstream chatbots' responses.

We don't store original files, only extracted text. We use secure servers in the US. We don't share or sell your data with anyone.