Codex May Be Quietly Chewing Through Your SSD

OpenAI Codex is supposed to help you code faster. For some users, it may also be making their SSD work far harder than expected.

Developers have reported that Codex CLI can write huge amounts of diagnostic data to local drives. One developer, Rui Fan from the Apache Flink project, said his main SSD recorded about 37 TB of writes in just 21 days of continuous Codex use. Stretch that over a year, and you get roughly 640 TB of writes.

This is significant because many commercial SSDs that have a capacity of 1 TB usually have a lifespan of about 600 TBW. This means that extensive use of Codex may lead to exhaustion of the write life span of your hard drive in a short while.

The Logging Problem That Is Out In the Open

The problem seems to arise due to the local SQLite logging of information in ~/.codex/logs_2.sqlite. It has been noted that after the software was updated in February, the logging level was raised to trace, which includes WebSocket communication and file operations.

The tricky part is that the log file may not look huge. SQLite can keep inserting and deleting records while still causing serious physical writes. That creates write amplification, where your SSD absorbs far more activity than the visible file size suggests.

For now, Linux and macOS users can redirect the log file to /tmp/ as a temporary workaround. Windows users have fewer simple options, so drive health monitoring with tools like CrystalDiskInfo becomes especially important.

Our opinion

This problem reminds us that AI tools do not only consume tokens, memory, and battery. They can also quietly consume hardware lifespan. If you rely on Codex every day, check your SSD write totals and keep an eye on official fixes.

Share this article with other developers, follow us on social media for more AI updates, and read our latest stories to stay ahead of problems before they hit your machine.