A developer has warned that Trae IDE, ByteDance’s competitor for Cursor, Copilot, or other AI-powered coding tools, collects excessive user data and sends it to the Chinese company’s servers, raising privacy and security concerns. ByteDance explained that some third-party extensions do not respect telemetry toggle.
An analysis of Trae IDE’s performance and telemetry practices is gaining significant traction on the Hacker News forum.
A developer using the GitHub handle segmentationf4u1t (segmentationfault) has revealed that Trae uses significantly more resources than Cursor and VS Code. Trae is built as a fork of VS Code.
Despite disabling telemetry in the settings, the tool tracks user activity and collects and transmits detailed information, including hardware specs, OS and architecture details, usage patterns, unique identifiers, project information, and more.
When the developer raised concerns on Trae's Discord server, their account was muted.
“I got spanked with a gag-hammer,” the developer said, noting that soon after the discussion, the word “track” was added to an automated blacklist, triggering an instant 7-day mute.
“Legitimate security concerns were treated as disruptive behavior,” the developer said.
Hundreds of reactions to the raised concerns have been posted on Hacker News.
Updated on August 1: ByteDance told Cybernews they’ve updated the platform with clarifications on how the telemetry toggle works—some third-party extensions do not respect the settings.
“The telemetry toggle mentioned by the user only controls telemetry data collected via VS Code. Telemetry data collected from other sources is unaffected by this setting,” the comment reads.
Now the platform’s telemetry toggle switch includes explanation that it only controls first-party extension telemetry, and participating third-party extension telemetry.
“Some third-party extensions might not respect this setting. Consult the specific extension's documentation to be sure. Telemetry helps us better understand how Trae CN - Dev is performing, where improvements need to be made, and how features are being used,” the UI reads.
The company also explained why the GitHub user received a time-out on Discord: the system flagged the word “tokens” due to crypto topics moderation.
The original blog author updated the post, saying recent Trae updates now explicitly acknowledge the presence of telemetry mechanisms, marking a shift toward greater transparency.
“I also overheard that a dedicated Privacy Mode is scheduled for release somewhere near August,” segmentationf4u1t writes.
“After direct contact with the Trae team, I can now confidently clarify — this was not censorship, but a poorly executed moderation mechanism. When I shared my findings in the official community Discord, my messages were auto-flagged due to a string match on the word “tokens,” which seems to be part of an internal keyword blacklist.”
The Chinese tech firm launched an AI-powered coding tool on macOS and Windows as a free alternative to Cursor or other AI-powered coding assistants. It features GPT-4o and Claude-3.5-Sonnet for free use as well as DeepSeek and other offerings.
What did the analysis previously claimed?
The analysis initially found that Trae, a fork of VSCode, runs 33 processes compared to VSCode’s nine processes, and transmits persistent telemetry despite user settings.
On a testing system, Trae consumed 6.3 times more RAM (5.7GB) than the standard VSCode (0.9GB) and three times more memory compared to Cursor (1.9GB RAM usage).
This issue was quickly addressed in a 2.0.2 release, which reduced the number of processes by twenty to around 13 processes using 2.5GB of RAM.
However, persistent network monitoring and outbound connections are “where the fun begins,” according to the developer.
Disabling telemetry in ByteDance's VSCode fork increases data sent to its servers
undefined CookingCodes (@CookingCodes) July 27, 2025
research here: https://t.co/IrLo4Bwxx4
Trae, allegedly, constantly connects to three endpoints on byteoversea[.]com top-level domain.
“I attempted to disable telemetry through the standard settings interface,” the developer writes.
“Disabling telemetry did not reduce network activity.”
The opposite happened, and the frequency of data collection increased.
“While actively using the editor, I observed around 500 calls within ~7 minutes, totaling up to 26MB of data transferred in that short timeframe,” segmentationf4u1t said.
As proof, the developer shared screenshots of network activity and intercepted payloads.
One data packet contains multiple persistent user identifiers, such as user_id, device_id, machine_id, biz_user_id, and user_unique_id, as well as detailed hardware fingerprints, such as the specific CPU model, exact RAM amount, and motherboard manufacturer.
The second shared telemetry payload captures more invasive real-time surveillance of user activity within Trae. It records precise timing when the user is active and what files they are editing, including full file system paths exposing the username, detailed behavioral metrics like mouse/keyboard activity, window focus states, or editor visibility, combined with detailed hardware specifications.
The developer warns that information is routed to ByteDance despite user preferences, and multiple persistent identifiers enable long-term tracking.
Updated on August 1st [09:20 a.m. GMT] with a statement from ByteDance and additional information.
Your email address will not be published. Required fields are markedmarked