AI answers in Outline

The OpenAI integration will allow for various AI functionality around the Outline workspace. Today, it enables “AI answers” to questions by semantically indexing your knowledge, retrieving related documents in realtime, and generating direct answers to questions using their content.

Setup

OpenAI

• Sign up for an OpenAI account, create an API key and add it to the server environment under OPENAI_API_KEY.

• You will need a database with the pgvector extension installed to store the search index. This can be a new database, or the same database you use for other Outline data. If you’re using a separate database then an additional DATABASE_URL_VECTOR environment config should be provided to tell Outline where to find this database.

• Restart the application server, Under Settings → Features → AI Answers you can now enable the integration which will automatically begin indexing content.

Azure AI Foundry

• Create a new Resource in Azure AI Foundry. Copy the API key and endpoint.

• Under “Model catalog” in the left side nav make sure to deploy both gtp-4o-mini and text-embedding-ada-002 models with their default names.

• Set the following environment variables in Outline:

  • OPENAI_URL=https://<your deployment>.openai.azure.com/openai/

  • OPENAI_API_KEY=<your API key>

  • OPENAI_API_VERSION=2024-10-21

• You will need a database with the pgvector extension installed to store the search index. This can be a new database, or the same database you use for other Outline data. If you’re using a separate database then an additional DATABASE_URL_VECTOR environment config should be provided to tell Outline where to find this database.

• Restart the application server, Under Settings → Features → AI Answers you can now enable the integration which will automatically begin indexing content.

Privacy and limitations

• AI answers are limited to information contained in documents that the logged in user has access to.

• The initial indexing may take a few seconds or a few minutes depending on your OpenAI rate limit and the number of documents in your knowledge base.

• Content inside attachments will not yet be indexed.