I built ctx to give agents a way to do a fast local search over previous session transcripts.

But I wanted to also give them a way to immediately surface "which agent wrote this exact line of code".

So you can start with a commit, a file, a line, or a PR, and then deterministically surface the exact tool call that wrote it, and the full agent transcript from that session.

It's really useful when you want your agents to know why code was written the way it was. It's kind of like a "super comment".

A normal code comment does leave some information about why code is the way it is.

But if you have selective access to the full original transcript, you can see everything about the code's provenance (like what tests were run, what was the reason it was there in the first place, etc)

This type of information would be way too verbose for a human to find useful. But agents can just ingest this stuff and make great use of it.

I like to instruct the agents to delegate this type of research work to cheaper high-throughput subagents like Luna for example.