Let the AI agent read exact source ranges, not just retrieve relevant chunks
complete
s
sere.loyalty_2s
When I ask the agent to read a specific chapter or section, it often retrieves only chunks of the content. It then makes several additional searches, trying different queries to piece together the rest.
This is slow, uses unnecessary tool calls, and makes it difficult to know whether the agent has actually read the entire section. Relevant search results are useful for finding information, but they are not a reliable substitute for reading a continuous passage, especially when context, qualifications, or counterarguments matter.
Requested behavior:
- Give the agent a dedicated tool to read a specific source range, such as a chapter, section, PDF page range, or transcript timestamp range.
- Return the actual source content in its original order, rather than only the chunks ranked as relevant to a query.
- When a range is too long for one response, support sequential reading with pagination or a continuation mechanism. Clearly indicate what has been returned, what remains, and whether anything is unavailable or truncated.
For example, “Read chapter 3” should let the agent work through that chapter directly, rather than repeatedly guessing search terms to recover its contents.
Log In

Sankari Nair
updated the status to
complete
Thank you for writing this up so clearly. Chat can now read whole cards from start to finish, and connected apps can read long cards page by page, so the agent no longer has to keep guessing search terms to rebuild a section. Reading a specific chapter or timestamp range isn’t there yet, so if that’s still important to you, let us know how you’d use it.