Feat: Load existing Codex threads available via `codex resume`

[flamboh]

It'd be nice to resume existing threads from pre-t3-code. Loading them all doesn't seem right, so I'd suggest an 'import thread' option when right clicking a project. Input a thread id or select from a list and it adds it.

"Add the ability to import existing Codex threads by thread id. This should be an option on the project right click drop down menu. Use a dialog to show an input for thread ID and a list of recent threads not in the sidebar already."

[timm-u]

Additionally, it would be great if I could import threads from the Codex app into T3 Code, rather than just resuming them. It feels counterintuitive to start from scratch when I already have an extensive conversation history.

[siraht]

100% - this is crucial if I'm going to switch to T3 as my daily driver

[thorizer]

can't migrate to t3 without resume (too many existing sessions )

[tomocrafter]

I also would love to see this feature with claude code

[dtown123]

This app would pretty much be the only thing I ever use if I could resume my existing claude/codex sessions like I can in IDEs/console

[0xpaperhead]

I was initially working on just creating that implementation before starting to see other similar issues (such as this one) are already created and it seems like contribution is tightened (for obvious reasons previously outlined by Theo)

I think both providers persist sessions on disk in a stable location, as I reseached:

• Claude Code: ~/.claude/projects/<encoded_cwd>/<uuid>.jsonl
• Codex: ~/.codex/sessions/<YYYY>/<MM>/<DD>/<uuid>.jsonl

Sessions should also be project scoped.

regardless, both also expose native resume via claude --resume <id> and codex resume <id>. The data and the entry point already exist.

• Layers/ProviderSessionDirectory.ts abstracts session storage on disk.
• Layers/ProviderSessionReaper.ts manages session lifecycle.
• Layers/ClaudeAdapter.ts and Layers/CodexAdapter.ts are the seams for each provider.

[0xpaperhead]

I made it in my fork

[dtown123]

I made it in my fork

YES ALMOST THIS EXACTLY. That list should literally appear under the project folder in the side nav instead of the current managed chats. Right where it says "No threads yet"

[0xpaperhead]

yep! exactly my change. I tested it and it works. Gonna push to my fork soon.

Also I find myself reusing certain prompts so I am also adding a canned prompts feature

I made it in my fork

YES ALMOST THIS EXACTLY. That list should literally appear under the project folder in the side nav instead of the current managed chats. Right where it says "No threads yet"

[dtown123]

Please do! One my biggest problems with the current "AI meta" is "prompt hunting" where you remember something you did a few weeks ago, but you don't remember -where- you did it.

Was it codex? cursor? oh wait, maybe it was claude in the home directory?

I love the direction of T3 code, but I don't want it to become another disparate bucket that I have to check. I would prefer it to be the opposite; a one stop shop for any and all of your AI interactions no matter where they began.

[0xpaperhead]

yep exactly!! hunting for sessions and where you mentioned what!

I was thinking of adding semantic search across sessions. I like t3code

btw im the dev behind agentgrid.sh :

demo: https://www.tella.tv/video/organize-your-ai-coding-workflow-with-agent-grid-42t5

I even added a master-worker interface within agentgrid where you can talk to any of the workers but also let the orchestrator run your agentic engineering loop (build -> review -> audit -> loop -> PR ->etc...)

[0xpaperhead]

@dtown123 it works!

[materemias]

definitely needed also for claude code