Antigravity Conversation Fix, rebuilding a sidebar index from files on disk
Fixes missing and unordered conversation history in Google Antigravity
At a glance
- What is it?
- Antigravity Conversation Fix is a Python script, shipped as a Windows executable, that reconstructs the Antigravity sidebar index from the conversation files still on disk after the index is corrupted. It handles multiple folder and database variants, derives missing titles from the conversations themselves, and refuses to run while Antigravity is open.
- Who is it for?
- Antigravity Conversation Fix suits anyone whose Antigravity conversations vanished from the sidebar while the files are still on disk, since it reconstructs the index rather than the content, and it suits Windows users who want a single file to run without setting up Python.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 59 days ago.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on October 22, 2026, and from our analysis. They are not legal advice.
Editorial analysis
The problem is a corrupt index, not missing conversations
The failure this tool addresses is specific. Antigravity stores conversation data in two places: conversation files with a .pb or .db extension, kept in the user profile, and a sidebar index, which is a SQLite database in the application data folder, one per Antigravity variant.
When the index gets corrupted, the conversations still exist on disk but do not show up in the sidebar. Nothing was deleted. The listing that points at the files is broken. The tool works by scanning the conversation files, sorting them by date, pulling titles from brain artifacts or creating concise titles from readable user requests, and writing a clean index back to the database.
That distinction is the whole reason a repair tool is possible at all. If the conversations were gone, there would be nothing to rebuild from.
The README lists the symptoms it targets, and the list reads like a bug tracker. Conversations missing from the sidebar, conversations in the wrong order, placeholder titles instead of real names, titles lost after previous fix attempts, workspace assignments stripped on a rebuild, lost workspace assignments from v1.0 damage, missing timestamps causing a wrong sort, remote workspaces not recognised, a renamed application folder not detected, conversations split across multiple folders after an upgrade, only one of two installed variants being fixed, a new .db conversation format not detected, placeholder titles despite usable conversation text, and recent .db messages disappearing from the sidebar later.
The last two are the most telling. They are not corruption at all. They are the ordinary consequences of an index that was correct when written and stale now, which is why the tool also refreshes updated timestamps on a re-run.
Running it on Windows takes a download and a double-click
The quick start is written for Windows users, and it is deliberately simple. The README says no Python or developer tools are required, just download, run, done.
The sequence is six steps. Close Antigravity completely, using File then Exit or killing it from Task Manager. Download the Antigravity_Conversation_Fix.exe from the releases page. Double-click it, and a terminal window opens. The tool scans your conversations, rebuilds the index and shows you the results. When prompted for workspace assignment you pick an option, with Enter or 1 auto-assigning workspaces from your brain files, described as recommended, and 2 auto-assigning first and then letting you assign any remaining conversations manually. Then open Antigravity, and the conversations are back, sorted by date.
The README adds that a PC restart is normally not required if Antigravity was fully closed, and that you should reboot only if the changes do not appear.
That first step is not advice, it is a requirement enforced in code. The changelog for v1.07 records it as a safety change: the tool now exits without scanning or writing when Antigravity is running, preventing the application from later overwriting the repaired index. If you close the app, run the tool, and then start the app, the order is what makes the repair stick.
The project is a Python codebase, with rebuild_conversations.py at the repository root and an Antigravity_Conversation_Fix.spec file, which is the PyInstaller specification used to build the executable. There is a run.bat for the Windows path, a tests/ directory, a RELEASE_NOTES.md, and the MIT licence at the root. The releases are versioned V1.05, V1.06 and V1.07, the newest from 2026-08-26.
One script, three platforms, and a lot of path detection
Underneath the Windows quick start is a cross-platform Python script, and the path handling is the substantial engineering in this project.
The README documents where the data lives on each operating system. On Windows the conversations are in %USERPROFILE%\.gemini\antigravity\ or antigravity-ide\, and the database is a state.vscdb under %APPDATA%\Antigravity IDE. On macOS the conversations are in ~/.gemini/antigravity/ and the database is under ~/Library/Application Support/Antigravity IDE. On Linux the conversations are again in ~/.gemini/antigravity/ and the database is under ~/.config/Antigravity IDE. On WSL the conversations are in the home directory and the database path is auto-resolved from the Windows %APPDATA% through /mnt/c/.
The variants multiply. The tool automatically detects all folder name variants, listing antigravity, antigravity-ide, antigravity-backup, Antigravity and Antigravity IDE, and merges conversations from all locations, with duplicates removed and the newest location winning. The changelog for v1.06 records this as multi-folder conversation merge, and notes that users who upgraded from v1.x to v2.x no longer lose conversations that were only in the old or backup folder. The same version added multi-database write, discovering all existing Antigravity databases and writing the rebuilt index to every one of them, so that having both the standalone agent and the IDE installed is handled.
The WSL path is its own category of difficulty. The changelog for v1.05 records native WSL path detection, replacing the need to copy files by hand, and the same version added auto-detection of Python 3 with a built-in fallback for the case where the python command fails on macOS or Linux. Also in v1.05 is auto-detection of both old and new paths, for the case where the Antigravity IDE folder was renamed, and support for the 2.x antigravity-ide data folder.
Each of these is a small compatibility fix, and together they are the difference between a script that works on the maintainer's machine and one that works on yours.
Title resolution, and the priority order behind it
The part of the tool that needs the most judgement is deciding what a conversation should be called, and the README lays out the priority explicitly.
Titles already in the database come first, described as canonical Antigravity titles, preserved across re-runs. The only exception is the exact raw-prompt titles written by the first v1.07 build, which are reformatted. Then come brain artifact .md headings, used for conversations not yet indexed. Third is a concise title derived from the first valid user prompt in a .db conversation. Fourth, for legacy .pb conversations, the first readable USER_EXPLICIT or USER_INPUT request in the brain transcript or overview. The fallback is Conversation (date) short-id.
The ordering is conservative by design, and the changelog shows why. An earlier build of v1.07 produced long raw prompts as titles, so the fix shortens them: a prompt-derived title uses a short first sentence when available, collapses web links to their hostname, and otherwise stops at 10 words or 60 characters. Titles written by that earlier build are migrated on the next run. Legacy .pb conversations can recover a title from readable brain logs, but encrypted .pb data without a readable transcript or overview still falls back to the date-and-ID form, which is an honest admission that some data cannot be recovered.
Two more fixes from the same version handle ordering. A .db is preferred over a .pb with the same ID inside one data folder, and malformed early prompt records no longer hide a later valid prompt. The last of those is the kind of detail that only surfaces after someone hits it in the field.
The output is marked so you can see what happened. The legend gives a marker per title source: a plus sign for a title extracted from a brain artifact, a tilde for a title preserved from the existing database, an equals sign for a title derived from a conversation's readable user request, a question mark for the fallback, and WS for workspace metadata preserved or recovered.
What it repairs, what it cannot, and what it will not do
The limits are worth stating plainly, because a repair tool that runs against your IDE's data deserves careful reading.
The tool does not recover deleted conversation files. It rebuilds an index from files that exist, so if the.pb or.db files are gone, there is nothing to scan. It also does not restore conversation content, only the listing, the ordering, the titles and the workspace assignments. The workspace metadata is preserved or recovered from brain artifacts rather than reconstructed from scratch, and the v1.07 entry states that when an existing.db conversation file is newer than its indexed update time, the tool refreshes only the sidebar's updated timestamp, leaving created time, workspace mapping and unknown metadata unchanged. That narrowness is deliberate and is the reason the repair is safe to run repeatedly.
The workspace auto-assignment deserves a note of its own. Option 1 in the quick start assigns workspaces from your brain files, and the README calls this recommended. Option 2 assigns first and then lets you correct the remainder by hand. If you rely on workspace assignments, the manual option is the safer default until you have read the code.
The other limit is scope. This is a one-time repair with a version history that shows a specific bug being fixed each time, which is honest and also means it is not a monitor. The v1.07 changelog includes a line about tests, and it lists what they cover: generated SQLite and protobuf fixtures for title extraction, malformed input, title priority, timestamp refresh, metadata preservation and the active-process abort. That is a reasonably thorough fixture set, and it is the right place to look if you want to know what the tool has actually been exercised against.
The project is MIT licensed, the last push was on 2026-08-26, and the releases V1.05, V1.06 and V1.07 are dated 2026-05-20, 2026-05-22 and 2026-08-26. If you are considering this for a shared or managed machine, note that the executable is built from a Python script in the repository, so verify the download before running it.
Editorial conclusion
Antigravity Conversation Fix suits anyone whose Antigravity conversations vanished from the sidebar while the files are still on disk, since it reconstructs the index rather than the content, and it suits Windows users who want a single file to run without setting up Python. It is the wrong tool if the conversation files themselves were deleted, because the tool reads them as the source of truth, and it is the wrong tool if you need ongoing protection rather than a one-time repair, since it corrects a corrupted index and does not watch for the next corruption. Before running it, close Antigravity completely, because the tool exits without scanning or writing when the application is running, and back up the state.vscdb file if you can, since the README does not describe an undo path for a partial run.
Frequently asked questions
What does Antigravity Conversation Fix actually repair?
It rebuilds the sidebar index, not the conversations. Antigravity keeps conversation files in the user profile and a SQLite index in the app data folder, and when the index is corrupted the files remain on disk but the conversations disappear from the sidebar. The tool scans the files and writes a clean index back.
How do I run the fix on Windows?
Close Antigravity completely, download Antigravity_Conversation_Fix.exe from the releases page, and double-click it. The tool scans, rebuilds the index and prompts for workspace assignment, where Enter or 1 auto-assigns from your brain files and 2 lets you assign the remainder manually. No Python install is required.
Why does the tool refuse to run while Antigravity is open?
The v1.07 changelog records it as a safety change: the tool exits without scanning or writing when Antigravity is running, preventing the application from later overwriting the repaired index. Closing the app first is what makes the repair stick.
Can it recover titles for conversations that have none?
Yes, through a priority order. It uses an existing database title first, then brain artifact .md headings, then a concise title derived from the first valid user prompt in a .db conversation, then the first readable USER_EXPLICIT or USER_INPUT request for legacy .pb conversations, and finally a Conversation (date) short-id fallback when nothing is readable.
Does it work on macOS, Linux and WSL?
The README documents data locations for Windows, macOS, Linux and WSL, and the v1.05 changelog records native WSL path detection plus auto-detection of Python 3 with a built-in fallback. The packaged executable described in the quick start is for Windows.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/futureisinpast-antigravity-conversation-fix)Was this useful?