OpenAI · 2026-06-22 · major
OpenAI Codex — SSD-burning SQLite log bug patched after 640 TB/year reports
OpenAI Codex CLI shipped two patches that cut about 85% of its SQLite log writes. Users had measured 37 TB written in 21 days, on track for 640 TB a year and full-drive SSD wear in months.
OpenAI cut about 85% of Codex CLI's local log writes after a SQLite sink at TRACE level was found burning through users' SSDs.
Key specs
| Measured writes | 37 TB in 21 days |
|---|---|
| Extrapolated | ~640 TB/year |
| Log volume cut | ~85% |
Quick facts
| Tool | OpenAI Codex CLI |
|---|---|
| Issue | #28224, opened 2026-06-14, closed 2026-06-22 |
| Reporter | @1996fanrui |
| Measured writes | ~37 TB in 21 days uptime |
| Extrapolated | ~640 TB/year on default settings |
| Fix | PRs #29432 and #29457, both merged 2026-06-22 |
| Reduction | ~85% of log volume removed |
What is it?
OpenAI Codex is the company's open-source coding agent that runs in your terminal. Issue #28224 reported that its local SQLite feedback log was writing roughly 37 TB in 21 days of normal use, on track for 640 TB a year. The maintainers closed the issue on June 22 after merging two fixes.
How does it work?
The log sink in the Codex CLI was installed with a default trace level, so it persisted every dependency target plus the raw WebSocket and SSE payloads of each model call. Two pull requests — #29432 stops logging every Responses WebSocket event, and #29457 filters noisy targets from the persistent log — together remove about 85% of the writes.
Why does it matter?
Consumer SSDs are rated for a few hundred terabytes of writes over their lifetime. At the measured rate, a Codex user on a 1 TB drive would burn through a full drive's endurance in months, not years. The fix means anyone running the latest Codex CLI keeps the agent's local history without trading their disk for it.
Who is it for?
Codex CLI users, especially heavy daily users on laptop SSDs
Frequently asked questions
- How much was Codex CLI writing to disk?
- The OpenAI Codex CLI was logging roughly 37 TB to a user's local SQLite store over 21 days of normal uptime. Extrapolated over a year that is about 640 TB of writes, which would exceed the rated endurance of a typical 1 TB consumer SSD many times over.
- What caused the SSD-fill bug in OpenAI Codex?
- Codex installed its SQLite feedback log with a default trace log level. That setting persisted every dependency target plus the raw WebSocket and SSE payloads from each model call, so the log file grew far faster than the agent's own activity would suggest.
- How did OpenAI fix the Codex logging bug?
- OpenAI merged two pull requests on June 22, 2026. PR #29432 stops Codex from logging every Responses WebSocket event, and PR #29457 filters out noisy log targets from the persistent store. Together they remove about 85% of the previous log volume.
- Do I need to update Codex CLI to avoid SSD damage?
- Yes. Users on the unpatched Codex CLI continue to write trace-level logs to SQLite at the same rate. Pulling the latest main branch and rebuilding picks up both fixes, after which the local log grows at a small fraction of the prior rate.
- Does the Codex bug affect ChatGPT or only the CLI?
- The bug is in the openai/codex command-line tool's local SQLite logging path, not ChatGPT. ChatGPT and other hosted OpenAI products do not write logs to your machine, so this disk-wear problem is specific to people running the Codex CLI on their own hardware.
Try it
git pull && cargo build --release # inside openai/codex