What we shipped, week by week. Rolling release notes straight from main. No marketing-language rewrites.
Per-service deploy scripts and a new account-pinned orchestrator land so every backend service can be shipped independently or in a single coordinated sweep. Dev configs are repointed at the new account. No user-facing changes.
End-to-end share-action gating lands with elevated-permission checks on every sensitive action (export, delete, and re-run) now all routed through a role verification. Shared report viewer gets a layout pass and the report cards are polished.
Three flagship features land on the same day. The AI Report Assistant brings natural-language analytics and schema-aware BI suggestions to Report Builder, powered by AI inference (Llama 3.3 70B). Report Re-Run lets you refresh any report against new data with one click. Password-protected share links add a bcrypt-hashed password + expiry timestamp layer to every public share URL.
Four back-to-back fixes close out the remaining share-access edge cases: a URL-key change, header handling for API calls, and an auth-refresh repair sweep. Report mail delivery is fixed and wrangler.toml gets a cleanup pass across services.
Shared-connection pagination, role assignment, and invitation flows all get fixes. Saved queries are stabilised (collection name + save flow), JSON export and object storage exports are repaired, and file-service export headers are handled for the new S3-compatible path.
Shell and cached-response paths get a performance refactor. Py-worker routing is updated. The big win of the week is refresh-token stability. MongoQUI used to fail after days of idle due to a stale refresh token; now the refresh flow is cached, scoped out of login, and wired through py-backend.
The architecture release. Py-backend moves into a cloud container, which unlocks per-user state, warm starts, and cheaper long-lived connections. The control plane migrates to our own database. Internal service-secret enforcement is now mandatory across service bindings.
Shell connection and query paths get an optimisation pass: faster first-query latency, fewer stale sockets. Detailed developer documentation ships for py-backend and internal integration.
Every push to main becomes a rolling build. Big feature blocks get their own dated entry. Marketing-free. Just what shipped.
Auto-updates from main. Your install gets every release in this log as soon as it ships.