A Codex SQLite logging issue puts local SSD writes under scrutiny

A GitHub issue about OpenAI Codex drew Hacker News attention after the reporter claimed Codex was continuously writing large amounts of data to its local SQLite feedback log database. The report names ~/.codex/logs_2.sqlite and its WAL and SHM files. On the reporter's machine, roughly 37 TB of SSD writes appeared after about 21 days of uptime, which they extrapolated to about 640 TB per year.

The issue includes a breakdown of retained rows and log categories. It reports 681,774 retained rows and about 1,035.6 MiB of estimated retained log content. TRACE logs accounted for 70.7% of retained bytes in the sample, while mirrored telemetry-related categories made up much of the rest. The issue remains open in the GitHub tracker.

The broader concern is that AI coding tools now sit close to the developer's daily environment. They run beside editors and terminals, observe local context, call services, and keep state. That makes logging policy a product-quality issue, not just an implementation detail. Users need reasonable retention, visible controls, and a way to notice when a background process is writing far more than expected.

The HN discussion focused on temporary workarounds, log levels, and telemetry boundaries. Some commenters shared SQLite-trigger approaches to block inserts, while others pointed out that an open repository makes local patches possible. The numbers still come from a specific report and should be checked against each environment. The signal is still useful: coding agents need to be evaluated on local resource behavior as much as model output.