Long-running Codex sessions become progressively slower and more resource-intensive for Captain to follow. Each live update processes the complete transcript again, so its cost grows with the full session history rather than only the new activity.
A synthetic production-path benchmark measured:
- 1,000 lines: 10.6 ms and 12.8 MB allocated
- 10,000 lines: 114 ms and 133.6 MB allocated
- 25,003 lines: 277 ms and 333.8 MB allocated
Repeated refreshes can therefore cause increasing update latency and sustained CPU and garbage-collection pressure during long conversations. These measurements cover parsing and normalization only and exclude database work.
Long-running Codex sessions become progressively slower and more resource-intensive for Captain to follow. Each live update processes the complete transcript again, so its cost grows with the full session history rather than only the new activity.
A synthetic production-path benchmark measured:
Repeated refreshes can therefore cause increasing update latency and sustained CPU and garbage-collection pressure during long conversations. These measurements cover parsing and normalization only and exclude database work.