What we shipped.

AI issue triage (beta) — auto-label, auto-prioritize, suggest assignee

Drop an unstructured bug report into the new issue dialog and we'll suggest the right labels, the right priority, and the right assignee based on your team's recent triage patterns. Workspace-scoped — your data only trains on your workspace.

• •Suggestions appear inline with a one-click "accept" / Cmd-↵ to accept all
• •Auto-priority uses your team's past P0 → P3 calibration, not generic heuristics
• •Privacy: model runs in a dedicated single-tenant inference container, no cross-customer training
• •Toggle off per-workspace in Settings → AI

Cycle planning v2 — drag-to-commit, capacity hints, sprint goals

Big rework of cycle planning. Drag issues from the backlog onto the cycle and we show capacity hints based on past velocity. Set a sprint goal that surfaces in the daily standup view.

• •Capacity-aware drag (red if over-committed, green if under)
• •Sprint goal field — markdown-rendered in the standup view
• •Velocity baseline now uses a rolling 6-cycle median (was 3-cycle mean)

Performance: Mongo compound index for the "Active" view

The most common workspace view (Active issues grouped by status) now uses a compound index. p95 dropped from 850ms to 18ms in staging. Live in production now.

• •Affects /api/issues?status=in-progress + /api/issues?status=todo queries
• •No customer action needed — index built in the background

Slack: link unfurls now show priority + assignee inline

Slack link previews of RelyvFlow issues now include the priority pill, current status, and assignee avatar. Cleaner skim. Existing Slack integrations updated automatically.

Fix: Push notification deep-links open inbox instead of issue (mobile)

Embarrassing — our URL parser was stripping the hyphen from issue identifiers (every identifier has one). All mobile notifications now correctly deep-link to the source issue.

Webhooks v2: exponential backoff + dead-letter queue

Outbound webhook retries now use 30s → 2m → 10m → 1h → 6h → 24h → 24h backoff (7 attempts). Failures land in a per-webhook DLQ you can inspect and retry from Settings → Webhooks.

• •DLQ entries retained for 30 days
• •Manual retry from the UI; bulk retry via API
• •New webhook delivery analytics: success rate, p95 delivery latency

Security: SAML 2.0 SSO out of beta

Workspace-level SAML 2.0 SSO is now generally available on the Business plan. Tested against Okta, Azure AD, Google Workspace, Auth0, OneLogin.

• •Just-in-time provisioning + SCIM for member sync
• •Identity provider-initiated and service-provider-initiated flows
• •Session enforcement: configurable max session age

Improvement: Cycle burndown projected-end-date math

Off-by-one in the linear extrapolation. Projection was reading "ends Friday" when it should've read "ends Thursday." Fixed.

New: Saved views — share with the workspace, pin to sidebar

You can now save any filtered issue list as a named view. Pin to the sidebar, share with teammates, set as the default landing view for new members.

• •15+ filter fields composable in views
• •Group by status, priority, assignee, project, label, or none
• •Sort by updated, created, priority, or manual

Region pinning: EU + US data residency

Business workspaces can now pin their data to either the EU or US region. Mongo + S3 + processing all stay in-region. Migration path for existing workspaces — talk to support.