I opened a new monorepo one afternoon and the Indexing indicator simply did not move. After a while I asked the agent about a class under src/, and it told me the file didn't exist. The status said done. The answer said empty.
My first instinct was to flail. I restarted, deleted what looked like cache folders, reopened the project — and none of it helped, because I had never asked which kind of stuck this was: never started, stopped halfway, or finished but blind.
What follows is the order I use now. Count before you delete. Touching the app's own internal folders is my last move, and only after the numbers fail to explain things.
One note on scope: I couldn't find official documentation for Antigravity's internal directory layout or for any indexer-specific settings keys. So nothing below depends on those names. Everything works on your project directly.
First, sort the symptom into one of three
"Indexing is broken" covers three different problems. Watch your OS monitor (Activity Monitor on macOS) alongside the status bar.
| Symptom | What the OS monitor shows | Where to look first |
|---|---|---|
| A: no progress, or stuck at 0% | CPU and disk both nearly idle | App state, which folder is open, app version |
| B: progresses, then stops | CPU pinned, or disk reads that never end | Project too large, one bad file, symlink loops |
| C: "complete," but the agent can't see files | Quiet | Ignore rules, the folder level you opened, which tree is being read |
Once you know which one you have, jump to that section. The first two moves are the same for all three.
Two moves for every case: version, then reopen
Some indexing and freeze fixes arrive as app updates rather than settings. I used to check the version only after rewriting my ignore file, which was backwards.
I keep a table of freeze- and slowness-related fixes from the official changelog in Antigravity feels slow or freezes: isolate the cause before you fix it. If your symptom resembles a row there, update first and retry the same action.
If nothing changes, quit the app completely and open the same folder again. If that fixes it, you can stop reading.
Step 1: turn "big" into numbers
Symptoms A and B often come down to the project being larger than you assumed. Don't guess. Run this from the project root:
#!/usr/bin/env bash
# index-scope.sh — run from the project root (read-only)
set -u
echo "== files tracked by git"
git ls-files 2>/dev/null | wc -l
echo "== all files, excluding .git and node_modules"
find . \( -name .git -o -name node_modules \) -prune -o -type f -print | wc -l
echo "== files over 5MB (top 10)"
find . \( -name .git -o -name node_modules \) -prune -o -type f -size +5M \
-exec du -h {} + 2>/dev/null | sort -hr | head -10
echo "== symlinks (up to 20)"
find . \( -name .git -o -name node_modules \) -prune -o -type l -exec ls -l {} + 2>/dev/null | head -20
echo "== size of .git"
du -sh .git 2>/dev/nullThree ways to read the output:
- A big gap between tracked files and all files means untracked generated output or data is piling up. That gap is where your ignore candidates live.
- Files over 5MB are usually generated code, bundles, or data. The agent gets little from reading them, and they cost a lot to read.
- Symlinks that point to a parent directory, or back at themselves, are a loop suspect.
In my case it was an auto-generated GraphQL schema, a single file of several megabytes, sitting inside the scanned tree. Until I saw the number, I had been blaming settings.
Step 2: add ignores, then verify they work
Once you've found the heavy files, add them to .antigravityignore. Think of the syntax as .gitignore-style.
# .antigravityignore — at the project root
# Build output
build/
out/
*.bundle.js
*.min.js
# Generated code
**/*.generated.ts
**/openapi-types.ts
# Large data
*.csv
*.parquetReopen the app and watch the indexing behavior. One caution: writing a rule and having the rule take effect are two different things. I walked through the common pitfalls, checking each with a matcher, in Four places to look when .antigravityignore isn't working. Use that checklist right after you add rules.
Step 3: set .git aside and try once
If A or B survives a clean ignore file, .git itself may be the snag. A report on Google's developer forum describes a Language Server crash loop when the project root contains .git, with chat never initializing (Language Server crash-loop on projects with .git folder).
According to that report, the log shows a LS update error: Canceled line, and neither listing .git/ in .antigravityignore nor excluding it from file watching helped. Moving .git out reportedly let the server start cleanly. The reporter themself calls the cause a guess, and the setup was a Node.js project on Windows 11, so please read this as one environment's experience rather than a rule.
To test it, leave your real .git alone. Copy just the working tree somewhere else and open the copy:
# Copy without .git (your original project is untouched)
rsync -a --exclude '.git' ./ /tmp/index-probe/
# Open the copy in Antigravity and see whether indexing runs to the endIf the copy finishes, .git is involved, and adding your OS and app version to that forum thread would help others. If the copy stalls too, this isn't about .git; go back to the numbers from Step 1.
Symptom C: "complete," yet the agent can't find it
First confirm the file is actually outside the index. More often than not, the cause was not a bug but my own ignore rule, or the folder level I had open.
Check whether git considers the file ignored:
# If a rule matches, it prints which one (no output means no match)
git check-ignore -v src/services/user.tsgit check-ignore can't evaluate .antigravityignore, so use it for .gitignore and use the matcher approach from the article linked above for the Antigravity file.
Then check which folder you opened. Sometimes the parent is open while you ask about a child project, or a same-named folder from somewhere else is open instead. Compare the folder path in the window title with the path of the file you're asking about.
Finally, doubt the agent's answer for a moment:
# Confirm with your own eyes that the symbol exists
grep -rn "getUserById" src | headIf it exists and the agent still can't find it, return to Steps 1 through 3. If it doesn't exist, the problem isn't indexing; it's a misremembered name, the kind of slip that's easy to make when you switch branches a lot.
When nothing explains it
Some days the cause stays hidden. On those days I write down only what I could confirm, so that future me, or whoever reads a bug report, gets somewhere faster:
- Antigravity version and OS
- Output of the Step 1 script (file counts, names of the large files)
- Whether the copy without
.gitran to the end - Whether anything changed before and after adding ignore rules
With those four lines, a forum report or my own later review goes much faster.
Don't hand-delete the app's internals
One line I try to hold: I don't delete internal app directories whose behavior isn't documented. If deleting seems to fix things, I no longer know what actually worked, and the next time it happens I'm stuck deleting again.
Count, isolate, keep only what helped. Even when this doesn't fix the problem, it leaves a record of what it wasn't.
Your next step
Run the Step 1 script once in the project you have open right now. You only need to keep two of the numbers: tracked files and total files.
With those two in your notes, the next time Indexing stalls you'll already know where to look first. That's where I start, too.