Release v1.3.0 · cortexkit/openai-auth

New

Keeps working when the vault is down. In vault mode, the plugin now remembers the last token the vault gave each account, in memory only. When the vault can't be reached, requests keep using that token until the vault is back, instead of failing. The token is dropped as soon as the vault refuses the account, the account leaves the vault, you disconnect, or OpenAI rejects the token. Local accounts are still never used in vault mode. A process that starts while the vault is already down has no remembered token, so its requests fail until the vault is back. On OpenCode 1, OpenCode 2 and Pi.
No waiting during a vault outage. Once the plugin notices the vault is unreachable, each request uses the remembered token at once. When the vault comes back, or refuses an account, the plugin notices within about a second of the next request.
The vault and insula push changes. Instead of asking the vault for its account list every five seconds, the plugin now gets a notice when the list changes and asks only then, plus a safety check of the list every five minutes. Usage readings work the same way: insula sends a notice when it has a new reading, and the plugin reads it then instead of once a minute. With an older vault or insula that can't push, the plugin keeps checking as before.
Pi 1.3.0. The Pi package is tested with Pi 1.3.0.

Windows file saves. On Windows, each save of the plugin's account and settings files had half a second to finish. Checking that the plugin still held its lock on the file counted against that time, so on a slow machine (for example with antivirus scanning) every save failed before the file was written. The save is now always attempted.

No upgrade steps are needed.

Full Changelog: v1.2.0...v1.3.0