You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
docs: 0.1.0 upgrade notes for retention, encryption key and closed signup #152
Recordings now default to 30-day retention, down from 365. On upgrade, the next retention run deletes recordings older than 30 days. Set THINKER_RECORDING_RETENTION_DAYS=365 to keep the old behavior.
The REQUEST_RECORDING_* variables are renamed to THINKER_RECORDING_*.
From the #150 qa-docker run: pass THINKER_ENCRYPTION_KEY_PREVIOUS=${THINKER_ENCRYPTION_KEY_PREVIOUS:-} in the docker/docker-compose.yml environment and document it in docker/.env.example. Without it, key rotation needs a compose override.
From #29 (PR #149): the one-time setup token lives in the memory of each process, so on a multi-replica deploy the first setup must hit the replica whose log printed the token (or run with a single replica until the admin exists). Document this alongside the other multi-replica notes.
Activity
patrickbrandao commentedon Oct 14, 2026
Added from the #150 review (should-fix 1 and nits):
Document the upgrade recipe for installs that relied on the old
BETTER_AUTH_SECRETfallback:THINKER_ENCRYPTION_KEY.THINKER_ENCRYPTION_KEY_PREVIOUSto the oldBETTER_AUTH_SECRET.The recipe goes in DOCKER_README, the dev guide and the 0.1.0 notes, and the missing-key boot error should point to it.
Fix the
docker/install.shcomment that still mentions the fallback (lines 291-296).Mention this use of PREVIOUS in the
crypto.util.tsdoc comment.patrickbrandao commentedon Oct 14, 2026
From the #150 qa-docker run: pass
THINKER_ENCRYPTION_KEY_PREVIOUS=${THINKER_ENCRYPTION_KEY_PREVIOUS:-}in thedocker/docker-compose.ymlenvironment and document it indocker/.env.example. Without it, key rotation needs a compose override.patrickbrandao commentedon Oct 14, 2026
From #29 (PR #149): the one-time setup token lives in the memory of each process, so on a multi-replica deploy the first setup must hit the replica whose log printed the token (or run with a single replica until the admin exists). Document this alongside the other multi-replica notes.