[BUG] Conversation history lost after power outage - v1.20.5

Google Antigravity
1

All conversation history permanently lost from UI after power outage.
4 conversations (.pb files, 18.5 MB) are INTACT on disk but UI shows zero.

context_state/ directory is EMPTY (corrupted index).
trajectorySummaries in state.vscdb lost 3 of 5 conversation UUIDs.
No auto-recovery mechanism exists. Restart/reinstall does not help.

Ultra subscriber. This is unacceptable — days of accumulated context gone.

Fix: auto-rebuild index from .pb files if index is empty on startup.

Antigravity v1.20.5, Windows 10 Pro.

2

My God, are there people here who can fix this error? Chats aren’t saved when the system crashes, they seem to be completely deleted.

3

Same issue i experienced yesterday.

4

Oh my god, someone will solve this problem of months of context being destroyed. Work stopped with explaining contact code with thousands of lines of folders and files. After a crash, all dialogs are permanently destroyed. You pay for an Ultra subscription, and it’s literally the worst on the market.

5

i have a workaround i asked the agent to check the plans to grab it from antigravity brain folder but it cant decrypt .pb files any more unfortuently same thing happened and i reported this before many times

6

I don’t think this AGY devs can fix that.

7

Hello everyone!

Thank you for bringing the issue with conversation history to our attention. Our engineering team is currently investigating the matter, and we appreciate your patience as we work toward a resolution.

9

https://www.reddit.com/r/google_antigravity/comments/1rz0xkt/conversation_history_database_manager_view_edit/

10

What i found so forth:

The Root Cause

  1. The Trajectories Are Safe: All conversation .pb files are still saved locally under ~/.gemini/antigravity/conversations/.
  2. The SQLite WAL Panic: When Windows abruptly sleeps, reboots, or crashes, the SQLite index file ~/.gemini/antigravity/conversation_summaries.db-wal is left mid-transaction with an uncheckpointed lock.
  3. Destructive Reset: On startup, language_server.exe encounters the lock without a graceful retry/timeout and logs:
    summary store: reset due to error; will reconcile
    Instead of recovering the WAL or waiting, it drops and recreates an empty conversation_summaries.db.
  4. No Re-indexing: The server does not scan conversations/*.pb to repopulate the database, leaving the sidebar completely empty even though all data exists on disk.
  5. Windows Path Encoding Bug: Additionally, workspace_uris in conversation_summaries.db has an encoding discrepancy between URI-escaped drive letters (file:///d%3A/...) and unescaped (file:///d:/...), causing active workspace filters in the sidebar to miss sessions.)
  6. The process seems to be missing Safe SQLite Flags like PRAGMA busy_timeoutand any auto-reconciliation and or population

Environment Where Verified

  • Antigravity Version: v2.17.0 (and earlier v1.20+)
  • Process: language_server.exe
  • OS: Windows 10 / 11

How to Recover (Workaround for Users)

You can restore all your history by running a script to re-insert the metadata from your existing .pb files into conversation_summaries.db with both %3A and : workspace URI variants. Once re-inserted, the IDE sidebar immediately recognizes and reloads all conversations.