Release v0.59.0 · Morelitea/initiative

AI configuration is now connection-based with a single app-wide mode. An operator chooses whether AI providers are configured at the platform level (the operator's connections apply to every guild), per guild (each guild admin configures its own), or disabled — one mode app-wide, mirroring the sign-in posture. A connection defines the provider, model, and an optional shared key; guild members attach their own API key and pick which connection they use, and never set the destination themselves. Standalone per-user AI (outside a guild) has been removed. An existing platform or guild provider is migrated into a connection automatically and its key is preserved; per-user AI keys are not migrated.

AI provider credentials are now stored per member inside their guild's isolated database schema under row-level security, are never returned by the API (only whether a key is set), and are purged when a user is deleted or anonymized. A connection's destination (provider and base URL) is always set by the mode's owner, so a member's key is only ever sent to that owner-set destination. Every outbound AI call — subtask/description/summary generation, connection tests, and model listing — flows through the shared guarded egress that connects only to the policy-validated address; private or internal addresses are permitted only for an operator-configured local (Ollama) model, never for a guild admin or member.
Personal API keys are now stored behind database row-level security and are reachable only through the system engine — the request path holds no grant on the table at all — matching the server-side session store. Creating, listing, deleting, and authenticating with API keys is unchanged.

Downloads

No new app build this release — installed apps update automatically over the air on next launch.

docker pull morelitea/initiative:0.59.0

Or use latest tag. View on Docker Hub.