🚨CLAUDE CODE USERS: READ NOW!! YOUR SESSIONS EXPIRE IN 30 DAYS WITHOUT AUTO SAVE HOOK WIRED! -MillaJ

[milla-jovovich]

Urgent message from Milla-Jovovich: Important for Claude Code users who have sessions they thought were auto-saved without following README instructions.

Hey everyone! I wanted to reach out as quickly as possible after I saw some issues that users were having with the AUTO SAVE HOOKS not firing.

I'm sure most of you have been testing MemPalace before committing anything major, but I'd rather err on the side of caution and not chance anyone risking anything important. Especially as I see we have a lot of first time users coming in.

Though the README has all the information, the full instructions exist via a link to the mempalaceofficial.com website.

Maybe this has caused some confusion because this is where ALL instructions live, including the AUTO-SAVE HOOKS wiring instructions are.

If you have not gone through the README, clicked on the links for full instructions and wired the auto-save hooks, your info has not been saving to the palace.

This is important and urgent for Claude Code users as there is a strict 30-day deadline before your JSONL files will be deleted.

There is no time limit for Codex, but again, if you haven't wired the hooks, the sessions have not been added to your palace.

Rather than bothering to wait on another update, I wanted people to have the most time they can to get their sessions filed, so here are some easy solutions you can implement immediately to make sure your information can still be saved.

Just remember: if you have any sessions in Claude Code that you want saved into MemPalace, you must act now — Claude Code only retains the JSONL transcripts for 30 days. After that they are auto-deleted. The clock is ticking FYI!

Here is a list of manual ways to get the info you want or need right now.

---

Get your info out of Claude Code right now:

Four small standalone tools. Read-only on your data. No MemPalace required for tools 1–3.

1. Back up everything still in ~/.claude/ — get your active sessions OFF the 30-day clock. Rsyncs your transcripts to ~/Documents/Claude_JSONL_Backup/. Idempotent, safe to re-run.

curl -fsSL https://gist.githubusercontent.com/milla-jovovich/25b9d43b3518674c80d6e7e54c499362/raw/backup_claude_jsonls.sh -o backup.sh && chmod +x backup.sh && ./backup.sh

2. Find what already survived in your backups — scans iCloud, Dropbox, Google Drive, OneDrive, Documents, Desktop, and Volumes for orphan Claude Code transcripts you may have forgotten about. Shows date + size + topic preview per file so you can see what each session was about.

curl -fsSL https://gist.githubusercontent.com/milla-jovovich/25b9d43b3518674c80d6e7e54c499362/raw/find_orphan_claude_jsonls.sh -o find_orphan.sh && chmod +x find_orphan.sh && ./find_orphan.sh

3. Render any JSONL into readable text — strips tool calls and system noise, keeps user/assistant turns with timestamps. Useful when you want to actually READ a saved session.

curl -fsSL https://gist.githubusercontent.com/milla-jovovich/25b9d43b3518674c80d6e7e54c499362/raw/render_jsonl.py -o render.py && python3 render.py <path-to-jsonl> conversation.txt

4. Add a /save slash command — for ongoing capture. Drops into ~/.claude/commands/save.md so you can save any session manually with /save (requires pip install mempalace).

curl -fsSL https://gist.githubusercontent.com/milla-jovovich/25b9d43b3518674c80d6e7e54c499362/raw/save.md -o ~/.claude/commands/save.md

If MemPalace is already installed, also worth one shot to bring everything searchable:

mempalace mine ~/.claude/projects/ --mode convos --wing claude_imports

Codex users: same auto-save gap applies — your sessions aren't filed into the palace automatically unless you've wired hooks. No 30-day clock for you, but your ~/.codex/sessions/ and ~/.codex/history.jsonl files still aren't searchable until mined:

mempalace mine ~/.codex/sessions/ --mode convos --wing codex_imports

[MrSuddenJoy]

@milla-jovovich what are these files we are told to download with curl? How can we be sure they are not malicious ones? Why this pertains to Claude Code users only? What about users of other tools like ChatGPT, Gemini etc? If this comes from Claude itself, than where is the link to this being introduced by Claude's developers blog or similar?

Oh, and btw, Im still waiting for the proof that you indeed are that Milla, actress.

[LisandraMoura]

https://www.instagram.com/p/DWzNnqwD2Lu/

[mvalentsev]

@MrSuddenJoy on the identity question, this post may help: https://www.instagram.com/p/DXzuvBJSHuL/.
btw, MemPalace contributors in the comments.
See also #1390 / #1391

[milla-jovovich]

@MrSuddenJoy dude I made them fast ok? they were for a time sensitive situation, do you really think I'm going to put "malicious code" on your computer? like why? and did you even read the post? because if you didn't wire the auto save hooks your sessions aren't being saved and Claude has a 1 month time limit before they delete the json from their servers. so chill.

[MrSuddenJoy]

do you really think I'm going to put "malicious code" on your computer?

@milla-jovovich Be it you, be it someone else; does not matter single bit...... rule of thumb when dealing with links is NOT to trust them (or, at the very least, use with caution). This is especially true for production env's. Better to feel safe than to have to set (complex) infra from scratch because of some malicious software got from link.

like why?

@milla-jovovich Same story every time, just played out differently: to STEAL one's data/money/etc.

did you even read the post?

@milla-jovovich yes, Ive read it. What I also have done is I'd gone to Claude's site and nowhere did I find info about 30 days' retension of JSON.

[igorls]

@MrSuddenJoy — fair concerns, here's what I can offer as a code owner:

On script trust: the four tools are now merged into the org repo via PR #1391, reviewed in a public diff. The install URLs in my comment above are pinned to the merge commit SHA — those are immutable; the contents at that path can never change. For any script from any source, the safer flow is git clone and read before running, which works fine here.

On the 30-day retention: documented at https://docs.claude.com/en/docs/claude-code/data-usage. The setting is cleanupPeriodDays in your Claude Code settings.json, default 30. It governs local cleanup of ~/.claude/projects/*.jsonl on your own machine — not a server-side delete. You can raise the value if you want, but the default applies to most installs, which is why the timing matters.

On Claude-specific: the 30-day cleanup is specific to Claude Code's local storage behavior. ChatGPT and Gemini don't store JSONL transcripts on your machine the same way, so this particular backup tooling doesn't apply to them. MemPalace handles other transcript sources (e.g. Codex) separately via mempalace mine — this discussion is just the Claude-Code-specific cliff.

Hope that helps.

[igorls]

Update — PR #1391 is merged. The four tools now live in tools/ on develop. For anyone who prefers a pinned commit URL (immutable, reviewed in PR), here are the same install steps from the org repo at merge commit 435f0ad:

# 1. Backup all Claude Code JSONLs (read-only on source)
curl -fsSL https://raw.githubusercontent.com/MemPalace/mempalace/435f0ad348dedab264dacb727f9b390a85163a00/tools/backup_claude_jsonls.sh -o backup.sh \
  && chmod +x backup.sh && ./backup.sh

# 2. Find orphan transcripts in iCloud / Dropbox / Drive / Documents / etc.
curl -fsSL https://raw.githubusercontent.com/MemPalace/mempalace/435f0ad348dedab264dacb727f9b390a85163a00/tools/find_orphan_claude_jsonls.sh -o find_orphan.sh \
  && chmod +x find_orphan.sh && ./find_orphan.sh

# 3. Render any JSONL transcript to readable plain text
curl -fsSL https://raw.githubusercontent.com/MemPalace/mempalace/435f0ad348dedab264dacb727f9b390a85163a00/tools/render_jsonl.py -o render.py
python3 render.py <path-to-jsonl> conversation.txt

# 4. Install the /save slash command
curl -fsSL https://raw.githubusercontent.com/MemPalace/mempalace/435f0ad348dedab264dacb727f9b390a85163a00/tools/save.md \
  -o ~/.claude/commands/save.md

Or git clone https://github.com/MemPalace/mempalace and run them from tools/ directly.

Once your transcripts are backed up, file them into your palace:

mempalace mine ~/.claude/projects/ --mode convos --wing claude_imports

Note on the 30-day window: that's Claude Code's default for cleanupPeriodDays, configurable in your settings if you want to extend it — but the default is what almost everyone has, so the timing still applies. Back up first, configure second.

Thanks @milla-jovovich for shipping this on the right day.