Key findings
- ≈ 26 MBof data in a single session
- 500+telemetry requests
- ≈ 7 minof ordinary editor use
- 53,606 Blargest single batch
- Opt-outEven with opt-out enabled, Trae sends data from six categories.
- DataThe packets contain detailed hardware metadata, information about activity in the editor and unique identifiers of the workspace, machine and user.
- EndpointsConnections to key ByteDance endpoints persist, and the results are reproducible.
- ResponseThe Trae team got in touch; a privacy mode (around August) and greater transparency have been announced.
Scale
The footprint of a single session
In about seven minutes of ordinary IDE use we recorded more than 500 telemetry requests, weighing roughly 26 MB in total. The single largest packet was 53,606 bytes.
Telemetry requests in a single session
- 1 marker = 1 request
- 1 block = 100 requests
Scope
What ends up in the telemetry
Even with opt-out enabled, Trae sends data from six categories.
Data categories in the packets
- System
- operating system, CPU, RAM, architecture, screen resolution
- Device identifiers
- unique user, machine and session IDs
- Usage data
- file types, open files, session time, focus state
- Performance metrics
- response time, error status, SDK version
- Location and language
- region, language, time zone offset
- Workspace
- project identifiers, recent files (obfuscated)
Findings
What exactly we checked
Context
Trae is a fork of Visual Studio Code (Code OSS) developed by ByteDance. It inherits most of the base editor’s mechanisms – including the telemetry configuration. In newer versions the interface openly acknowledges the presence of telemetry, which is a step towards greater transparency.
Method
We captured the traffic with standard network monitoring tools; the results are reproducible. Connections to key ByteDance endpoints persist:
ByteDance endpoints
mon-va.byteoversea.commaliva-mcs.byteoversea.common-va.byteoversea.com/monitor_browser/collect/batch/?biz_id=marscode_nativeide_us
Data scope
The packets contain detailed hardware metadata and information about activity in the editor: the language of the active file, working time, window state and unique identifiers of the workspace, machine and user.
Note
Some of this data is sent even after telemetry is switched off.
Update
What changed between MK1 and MK2
The report was produced in two instalments. The second verifies the earlier findings and corrects one misunderstanding.
- MK1
First findings
- Analysis of the VS Code fork’s telemetry and observation of its network traffic.
- A community moderation thread: automatic flagging of messages.
- MK2
Verification and contact with the team
- Confirmation of persistent connections to ByteDance endpoints.
- The ‘automatic censorship’ turned out to be misconfigured moderation – a match on the word ‘tokens’, not a deliberate action.
- The Trae team got in touch; a privacy mode (around August) and greater transparency have been announced.
Methodology
Method and limitations
- Method
- We captured the traffic with standard network monitoring tools; the results are reproducible.
- Sample
- A single session: about seven minutes of ordinary editor use.
- Classification
- The data categories are a classification of what is found in the packets, not their shares.
- Correction
- In MK2 we corrected one misunderstanding: the ‘automatic censorship’ turned out to be misconfigured moderation, not a deliberate action.
Sources
Sources and citations
Media citations
How to cite
TwojeSystemy, ‘Telemetry and privacy in Trae IDE’, MK1 24 July 2025, MK2 31 July 2025. https://twojesystemy.com/en/publications/trae-telemetry-privacy/