Hi everyone, here is Jean-Louis from Germany (but French 
)! This is my first post in this forum and I look forward to your answers
What I’m trying to do
I am in search of a sustainable setup for syncing a vault across a MacBook, an iPhone and an iPad while keeping it accessible for chats with my preferred AI (Claude) over its different apps in these environments - ideally even read/write (“create a summary of all notes about xxx”, etc..).
I am already using Obsidian Sync for having my vaults in all environments. Access over Claude Desktop in MacOS is a no brainer. The complication arises with the access using Claude apps on iOS and iPad which do not have connectors to the local vaults in these environments.
Things I have tried
I am giving a try at putting my MacOS vault on Google Drive, which makes it available to the Claude apps in iOS/iPadOS over the Anthropic Google Drive connector. But I am well aware of the conflict I produce between the two sync services on my MacBook (Google Drive and Obsidian Sync). I read the post “Sync conflicts with Google Drive + Obsidian Sync?”
Does anyone have any (better) idea or an alternative setup for that goal? Please let me know!
Many thanks in advance. Best regards from the Rhineland!
Your instinct is right. It is worth explaining why, because the damage does not look like a typical sync conflict at first.
Two things are happening at the same time. Google Drive on macOS uses File Provider, so files can be online-only placeholders that download only when accessed. But Obsidian expects a normal directory that it can read and write at any time.
On top of that, you have two independent services syncing the same files without knowing about each other. When they disagree, one resolves the conflict. Then the other sees this as a new change and resolves it again. This ping-pong effect creates duplicated notes and silently reverted edits. You cannot fix this just by tweaking settings.
The only fix that works for me is to stop using one folder for two sync services. You should split the roles:
- Keep the vault on Obsidian Sync only. This is your main copy. One service, no conflicts.
- Put a one-way copy on Google Drive. Let your MacBook be the only device that writes to it.
You can do this with a simple command:
rsync -a --delete \
--exclude '.obsidian/' --exclude '.trash/' --exclude '.git/' \
~/Vault/ ~/"Google Drive/My Drive/vault-readonly/"
You can run this from a launchd agent every 15 minutes or so. Nothing else writes to that folder, so there are no conflicts to resolve. You just get a copy that might be a few minutes old, which is perfectly fine for AI tasks.
It is very important to exclude the .obsidian folder, and not just to save space. Plugins keep their settings in .obsidian/plugins//data.json and rewrite them constantly in the background. Note conflicts are visible (you get a conflicted file and fix it easily). But data.json conflicts happen silently. One machine wins, you get no warning, and weeks later your settings are a mess. Claude does not need these files anyway.
The read/write part is harder. If you let Claude write back to that Drive folder, you have two writers again and the same old problems. Instead, give writes their own separate lane. Let Claude create files only in a folder like vault-readonly/_inbox/. Exclude this specific directory from the rsync --delete command. Then, just pull files from it into your actual vault when you are on your Mac. This way, the two sync directions never touch the same files.
The .obsidian/data.json point is the detail most people miss until it bites them, good catch. Silent config conflicts are worse than the note conflicts because nothing tells you it happened.
One variant worth considering on the sync side, if you’re comfortable with git: instead of syncing through Google Drive, push the vault to a private git remote and have Claude’s side pull from a plain clone. Same one-way safety (Claude never touches the source of truth directly), plus two things rsync alone doesn’t give you: a full history you can diff and revert, and a natural merge point if you ever do want writes to flow back.
For the write-back lane specifically, instead of (or alongside) the manual _inbox/ folder, a small script that commits and pushes on a schedule turns “Claude drops files somewhere” into “Claude drops files somewhere, and there’s a timestamped, revertable record of every change it made.” That matters more than it sounds like once a model is actually writing into notes you care about: mistakes are recoverable instead of silent.
Running something close to this myself (VPS as the Claude-side mirror, vault as source of truth on the Mac), and the git history has already saved me once when a batch edit went further than intended.