Peer-to-peer Mesh Sync
No server, no internet.
ObjectBox Sync clients synchronize directly with nearby devices, over Bluetooth or Wi-Fi.
Device to device.
Changes travel peer to peer – no cloud account, no central server, no connectivity.
Available now.
As a preview for Android and Apple – iOS and macOS.
No server to run
Devices discover each other and exchange changes directly. Nothing to deploy, no account to create, no bill that grows with traffic. A Sync Server stays optional – add one later for cloud backup or to bridge into MongoDB.
Conflicts are handled
Sync Precedence decides first, then the Sync Clock, and first received breaks the tie. You do not write merge logic.
Vector search in the same database
ObjectBox runs ANN vector search on the device. Local AI that keeps working when the network does not – and the results sync when devices meet.
What people build with it - examples
In consumer apps
A family task app that works anywhere, e.g. in the supermarket or on a plane. It can be used to get things done as a group – from preparing the house for Christmas to going grocery shopping together. Of course, P2P Mesh Sync can also be used for multiplayer games that can be played anywhere, even if no one has a data tariff (e.g. when on holiday, or because younger kids just shouldn’t have one). Bear in mind, these are just two examples – the opportunities are endless.
In the field (Physical AI)
A warehouse that keeps working when the network does not. Forklifts read the pallet tags, carry the data with them and sync with whatever device is in range; the gateway at the ramp pushes to the cloud only when there is one. Of course, P2P Mesh Sync can also be used in tunnels and parking garages, on offshore rigs, on construction sites – anywhere the network is the part you cannot rely on.
Platforms and versions
C and C++ are the layer underneath the other SDKs: the mesh API is there, the network transport is registered by a platform SDK or by your own integration. Python is not part of the preview yet.
What the preview does and does not do yet
Mesh Sync is a preview – the API can still change before the final release. The documented defaults today:
- Up to 3 simultaneous peer connections
- Discovery runs 30 seconds initially, then in phases of 15 seconds with 45-second pauses
- Transaction logs are kept for 8 hours (28,800 seconds)
- Batches: 100 KB soft cap, up to 1,000 logs per batch
- Devices are grouped by mesh ID – today that is the only grouping mechanism
Most of these are configurable. If one of them is in your way, tell us – that is what a preview is for.
ObjectBox Sync is licensed per project as an annual fee, in tiers by number of devices – no metering.
How this compares
& vector engine
(CRDT-based)
(Vec1 0.7)
(syncs to a Qdrant server, beta)
From public documentation, 4 October 2026. If something here is out of date, tell us and we will correct it.
More on-device vector databases compared: On-device vector databases in 2026