Users have reported that the OpenAI Codex desktop application consumes substantial local resources. Notable reports include:
- One user reported roughly 150 GB of network traffic in a single month — approximately equivalent to five consecutive days of 4K video streaming.
- A V2EX user logged 4.8 TB of SSD writes in one month while Codex was reportedly idle in the background.
Users attribute the behavior to how the app operates: the client maintains a persistent WebSocket connection to the cloud, a "cloud sandbox" shuttles code back and forth step by step, and additional features such as GitHub synchronization and continuous background indexing appear to keep running without fully stopping.
Why this matters to users
Although the computationally heavy parts of agent operation run in OpenAI's cloud, the ongoing bandwidth usage and disk activity are borne by the end users' machines. Consequences highlighted by reports include:
- Increased bandwidth consumption, which can be problematic for users with limited internet plans.
- Accelerated SSD wear due to large volumes of writes.
- Continuous load on personal devices not designed for persistent background server-like workloads.
According to the accounts, some users have already migrated to fully cloud-hosted platforms to move the operational load off their personal hardware onto environments sized for that purpose.
Takeaway
The reported incidents suggest that the Codex desktop client shifts a portion of its operational costs — consistent network traffic and disk I/O — onto users' machines. The cited figures (about 150 GB of monthly traffic and 4.8 TB of monthly SSD writes in specific user reports) underscore that the local resource demand can be significant. Further measurements and explanations from service operators would be needed to determine the exact scope and root causes of the issue.
