AI Assistant overview

AI Assistant is an AI powered chatbot that will use data stored within your ThoughtFarmer intranet to help answer questions. It will not use data from your entire intranet, as feeding too much information into AI systems reduces their overall effectiveness. Instead, you can specify pages, sections, and documents to use as source material for AI Assistant.

We use a process called retrieval-augmented generation (RAG) to power AI Assistant. We use Anthropic's Claude Sonnet AI model hosted on AWS Bedrock. By hosting through AWS Bedrock all customer data is kept secure. Anthropic's Claude on AWS does not train based on your data, and does not retain any information that is sent to it.  

What does AI Assistant do?

• Uses natural language to get answers.
• Answers questions based on topics that you create. Topics feature specific content on your intranet. There is a limit to how much content you can include for the beta (1,000 pages).
• Answers questions based on content in the body copy of pages, rich text cards, quick links cards, or within the text of attachments stored within your intranet.
• Incorporates new / modified content in near real time so it will always answer based on the latest information.
• Provides links to content sources for people that want to read more.
• Users can give feedback on the accuracy of the answers.
• Tracks what people are asking and the answers that are provided.
• Fully secure. AI Assistant will never give answers for content an employee doesn’t have access to view.
• Fully secure. The AI model doesn't store or log your data.
• Complies with your data privacy requirements. The AI model never trains on your data.

AI Assistant works well for questions about policies, procedures, and manuals.

Future releases 

The following items are out of scope for the current version of the AI Assistant, but potentially may be included in future releases:

• Answering questions where the answer isn't found in the content of a page or the text of a document. E.g:
  • Who is the site administrator?
  • Who is the owner of the onboarding policy?
• Answering questions about people or teams 
  • “Who is on the Design team?”
  • “Who does Joan Lee report to?”
• Performing actions based on data within the system
  • “Reformat the following text following our voice and tone guidelines”
  • “Send this response to me”
• Accessing other data in the intranet
  • “What is the most recent company news blog post?”
  • “What were the most viewed pages in 2024?”
• Extrapolating on known data to make future predictions
  • “When will I get paid next?”
• Does not include content outside the intranet
  • No linked pages
  • No Sharepoint / Google Drive documents