"I restarted Antigravity and every conversation was gone." "Right after the update, my chat panel was just... empty." If you use Antigravity day-to-day, you've probably had this moment at least once. I still remember the morning I lost hours of carefully built context — the kind of sinking feeling that takes a while to shake off. This guide walks through how I diagnose whether the history is truly gone, how to extract what's still on disk, and the four habits I now follow so I never lose that much again. None of these steps require a paid plan or any extra tooling beyond what ships with most macOS, Linux, and Windows machines today.
Don't Panic Yet — Five Reasons Your History Looks Empty
When chat history "disappears," there's a real distinction worth making: data that has been physically deleted versus data that simply failed to load. In my own experience, missing-history situations almost always trace back to one of these five causes:
- Corrupted state files after a crash: If Antigravity exits mid-write — say, the OS forces a reboot or the process gets killed — the SQLite write-ahead log can end up inconsistent and the loader gives up on reading the history table at startup, even though the underlying rows are still intact
- Schema change after an update: Even minor version bumps occasionally tweak how chat sessions are stored, and old-format records sometimes fall out of the UI even though they're still on disk waiting to be migrated
- Workspace switching confusion: Antigravity scopes chat history per workspace. If you opened a different folder than the one you used yesterday, the panel naturally shows a different (often empty) history. This is by far the most common cause and the easiest to fix
- Automatic cleanup: When local storage usage crosses a configured threshold, older conversations may be pruned automatically. The default thresholds are conservative but not zero
- Sync collision: A second device starts a fresh, empty session and that empty state gets pushed up as the "latest" version, overwriting what was on your main machine. This is the failure mode that hurts the most because it can erase data on devices you weren't even using
The first thing I'd suggest is reopening the workspace you were using yesterday. If the chat reappears, you simply switched folders. If that doesn't bring it back, move on to the next step. Don't reinstall Antigravity yet — that decision should come last, not first, because a clean install often clears the very state file you'd want to recover from.
Step One: Confirm Whether Anything Is Still on Disk
Like most VS Code-based editors, Antigravity stores user state as a SQLite database under a globalStorage/ directory. The path differs by OS:
- macOS:
~/Library/Application Support/Antigravity/User/globalStorage/ - Windows:
%APPDATA%\Antigravity\User\globalStorage\ - Linux:
~/.config/Antigravity/User/globalStorage/
Start by inspecting the directory. We're looking for the size and modification time of state.vscdb — that's the headline indicator of whether your data still exists.
# macOS — list state files with sizes and modification times
ls -lah ~/Library/Application\ Support/Antigravity/User/globalStorage/
# Expected output:
# state.vscdb 5.2M <- main state DB (chat lives here)
# state.vscdb.backup 4.9M <- previous-run backup
# antigravity.chat/ <- extension storage for chatIf state.vscdb is several megabytes or more, your data is almost certainly still on disk. If it's 0 bytes or only state.vscdb.backup remains, the loader likely regenerated the database at startup and recovery becomes harder — but not impossible. In that case, the .backup file is your best lead, and you should copy it aside immediately before doing anything else, since some Antigravity startup paths overwrite it.
Step Two: Extract Conversations from the State File
state.vscdb is a plain SQLite database, so any system with sqlite3 installed can read it. Antigravity locks the file while running, so fully quit the app first before touching anything. On macOS, "Quit" from the menu bar is sufficient; on Linux, ensure no antigravity process remains by running pgrep antigravity and killing any stragglers.
# 1. Make a working copy — never operate on the original
cp ~/Library/Application\ Support/Antigravity/User/globalStorage/state.vscdb \
/tmp/antigravity_state_backup.vscdb
# 2. List chat-related keys, sorted by payload size
sqlite3 /tmp/antigravity_state_backup.vscdb \
"SELECT key, length(value) AS bytes FROM ItemTable \
WHERE key LIKE '%chat%' OR key LIKE '%history%' OR key LIKE '%conversation%' \
ORDER BY bytes DESC LIMIT 20;"
# Expected output (example):
# antigravity.chat.sessions|248531
# antigravity.recent.chats|45120
# workbench.panel.chat.viewState|9821Any key holding tens of kilobytes or more is likely the actual conversation log, stored as JSON. The next command exports it in a readable form.
# 3. Export the JSON value to a file you can open in any editor
sqlite3 /tmp/antigravity_state_backup.vscdb \
"SELECT value FROM ItemTable WHERE key = 'antigravity.chat.sessions';" \
| python3 -m json.tool > ~/antigravity_recovered_chats.json
# 4. Quick sanity check — read the first 50 lines
head -50 ~/antigravity_recovered_chats.jsonPretty-printing through python3 -m json.tool makes the file readable in any text editor — you can scroll through every prompt, every reply, and every timestamp. You won't get the original Antigravity UI back, but you can absolutely rescue the important exchanges and paste them somewhere safe, which is usually what matters most. If python3 isn't available, jq . produces equally readable output and is often already installed alongside development tooling.
Four Habits That Prevent the Next Loss
After going through this once, you will not want to do it again. Here are the four practices I now rely on. None of them require extra tools or paid features — they're just decisions you make once and then forget about.
1. Copy keepable conversations out the moment you have them
Long debugging sessions or hard-won design decisions get pasted into a project-local file (something like docs/decisions/ or _research/) immediately. Treating chat history as inherently volatile is, frankly, the saner mental model. Once you accept that the editor's chat panel is a working surface and not a permanent store, the rest of your habits fall into place naturally.
2. Use one-way Sync, not bidirectional
If you run Antigravity on multiple machines, two-way Sync is convenient but risky: depending on launch order, an empty session can be promoted to "latest" and overwrite real data. Picking one trusted device as the source and syncing in one direction eliminates almost all of these accidents. The trade-off is a slightly less seamless multi-machine experience, but the safety gain is enormous.
3. Snapshot the state directory nightly
I run the script below via launchd on macOS at 23:30 every night. The Linux equivalent is a cron entry, and on Windows you can use Task Scheduler with a .bat wrapper.
#!/bin/bash
# ~/bin/antigravity_snapshot.sh
SRC="$HOME/Library/Application Support/Antigravity/User/globalStorage"
DST="$HOME/Backups/antigravity"
mkdir -p "$DST"
# Keep only the last 30 days
find "$DST" -name "antigravity_*.tar.gz" -mtime +30 -delete
# Date-stamped archive
tar -czf "$DST/antigravity_$(date +%Y%m%d).tar.gz" -C "$(dirname "$SRC")" "globalStorage"
echo "Snapshot complete: $DST/antigravity_$(date +%Y%m%d).tar.gz"A few megabytes per night, retained for 30 days, gives you a rollback point for any recent date — a remarkable amount of peace of mind for the disk cost. If you're paranoid (which is reasonable here), point DST at an external drive or a cloud-synced folder.
4. Manual backup before every update
When a major update notification arrives, I tar up the state directory before applying it. If a schema change ever stops the new version from reading the old history, I always have one option left: roll back to the previous build, restore the snapshot, and decide on my own timeline whether to upgrade.
When Recovery Truly Fails
If everything above turns up nothing, the data is most likely physically gone. Two final places worth checking before you give up:
- OS-level backups — Time Machine on macOS, File History on Windows, or whatever snapshot tool your filesystem provides on Linux (Btrfs, ZFS, and so on). If any of these are running, you can usually restore a
state.vscdbfrom a few days ago without much trouble - Official support — depending on the account configuration, server-side traces of conversations may still exist. When you contact support, include the exact time the issue appeared, what changed just before it, the OS and version, and the affected workspace name. Vague reports get vague replies
For related troubleshooting, you may also find Fixing Antigravity Settings Sync That Won't Work and When Workspace Indexing Gets Stuck helpful — they cover adjacent failure modes in the same state-management subsystem and often provide context that helps explain what went wrong here.
One Thing to Do Right Now
If you plan to keep using Antigravity for a long time, it pays to assume history loss will happen eventually and prepare accordingly. Before closing this article, please do one small thing: copy your current state.vscdb somewhere safe. That single action expands your recovery options dramatically the next time something goes wrong. Full automation, nightly snapshots, support tickets — those can all wait. Just take that one step today, while the path and the steps are fresh in your mind.