Pi + ChatGPT stopped connecting—does fixing it mean giving up some security?
After the previous release tightened credential boundaries, Pi + ChatGPT hit a connection regression in the installed app. 0.1.162 restores the connection without solving it by giving Pi broader access.
Pi + ChatGPT connects again, and the fix does not weaken the credential isolation introduced in the previous release.
After the security boundaries were tightened, Pi + ChatGPT regressed in the installed app. A Turn could be created and then cancelled, making it look as if the credential never reached the Pi worker.
The problem was the Turn key, not the tighter security boundary
The root cause was that the main process used the wrong Turn key. As a result, the short-lived access credential was not routed back to the correct Pi worker.
0.1.162 fixes the dispatch of pi.credential.response, adds more accurate connection-progress signals for Pi, and adjusts the supervision timeout so the connection process has clearer state and more reasonable waiting boundaries.
The fix does not broaden credential access
The Worker environment remains narrowly scoped. Pi still does not receive refresh tokens, and the secret isolation introduced in 0.1.161 remains intact.
The fix is to deliver the short-lived credential to the correct Turn—not to make the app work again by exposing more credentials.
Packaged and installed smoke tests passed for Pi + ChatGPT, Codex, and DeepSeek. Full test suite: 379/379.
FinchStory