[FEATURE]

Open 💬 1 comment Opened Jun 9, 2026 by sh1boot

Preflight Checklist

  • [x] I have searched existing requests and this feature hasn't been requested yet
  • [x] This is a single feature request (not multiple features)

Problem Statement

Claud's summary:
The specific request worth filing: transcript files are stored as JSONL on disk but there's no in-session tool or UI affordance for searching them, and they get truncated from the context window during compaction. Both the user and the model lose access to earlier conversation content that may still be relevant. Useful capabilities would be: searchable transcript history in the web UI, and a tool the model can call to grep its own prior sessions.

Claude's not quite right about what I can access. I can scroll back past the last compaction where Claude says it cannot see, but it's slow and frustrating; and Claude could search that transcript more effectively than I can anyway.

Other issues similar to this one seem to focus on the human having access to the full transcript in the local tools. It looks like I already have this access via the web interface but I don't like it and I want Claude to have access to search it for me.

Proposed Solution

Record the complete transcript, across compactions, in a Claude-accessible location, so that I can ask Claude to search it for things we've already discussed, or ask it to deliver the file to me via an appropriate interface (ie., a download when using the web interface).

It should probably be a markdown file, because I did not enjoy trying to read the JSON file.

Alternative Solutions

Currently I just scroll and scroll and scroll in the web UI or the Android app. The android app lacks text search (so far as I know), so I just read fragments and hope to remember something as I scroll.

Priority

Medium - Would be very helpful

Feature Category

Other

Use Case Example

I'm working on a plan, and Claude is starting to suffer token-related dementia. I want to capture all of the fruitful operations we did so I can feed them to a fresh session. I have in mind one particular test which was very useful.

  • I can't remember how we did it.
  • Claude can't remember how we did it.
  • Claude actually suggested it in this same session, but has compacted it away.
  • Let's just look in the transcript and see what we did! 🎉
  • Claude can't see the transcript. ☹️
  • I can see the transcript but searching it is very difficult. 😢

Additional Context

_No response_

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗