If we add web search and url fetch capabilities to the assistant (see: #664 ), the assistant will be able to quote from external sources. However, the user will have no way of knowing when web search/url fetch actually succeeded, and where the assistant is accurately quoting from a source or writing an answer from memory.
To fix this, we can use the citations feature from the Anthropic API, and pass citations on to the front-end. They need to be wired properly through sub-agents so we don't lose the citation along the way, and design a way to display them within the assistant's conversation turn (perhaps with a collapsible sources section). The statuses sent to the front-end should reflect whether the assistant actually successfully read the pages or not.
If we add web search and url fetch capabilities to the assistant (see: #664 ), the assistant will be able to quote from external sources. However, the user will have no way of knowing when web search/url fetch actually succeeded, and where the assistant is accurately quoting from a source or writing an answer from memory.
To fix this, we can use the citations feature from the Anthropic API, and pass citations on to the front-end. They need to be wired properly through sub-agents so we don't lose the citation along the way, and design a way to display them within the assistant's conversation turn (perhaps with a collapsible sources section). The statuses sent to the front-end should reflect whether the assistant actually successfully read the pages or not.