Every feature slice that lands on main, written up when it goes live. No roadmap theater — if it's here, it's in the product.
Platform
The Approval-First Content Method, published
We wrote down the method the product is built on — how teams that publish under scrutiny should run AI content — and made it a public page. Client portals now also carry a quiet Contentelli attribution, so the reviewers your agency invites can find the workflow behind their review link.
The Method: six principles, from capture-not-prompt voice to edit-distance quality — at /method
This changelog: every feature slice that lands on main, written up as it ships
Portal attribution footer with first-touch source tracking
Product
Every review decision is now optimistic
A full polish pass over the core approval surfaces. Approvals, revision requests, scheduling, and stage transitions apply instantly and roll back cleanly if the server refuses — including compliance gate refusals. Loading, error, empty, and offline states are now real designed states on every surface, and the whole approver flow holds up at 375px.
Optimistic approve / request-changes on Review, Portal, and Calendar with exact rollback
Surface state matrix: skeletons instead of spinners, retryable inline errors, offline preservation
Mobile approver pass: the entire review loop works one-handed on a phone
Per-draft approval workflows for mixed-compliance queues
Portal
Client portal, rebuilt guest-first
Clients should not need an account to approve their own content. The portal now opens from a tokenized share link with no sign-in at all — reviewers add their name once and start approving. Magic-link sign-in, bulk approve, threaded comments, and a content calendar round out the review experience agencies hand to their clients.
Tokenized share links: send a URL, get approvals back — no client accounts
One-time magic-link sign-in for named client reviewers and contributors
Bulk approve with per-draft reconciliation, threaded comments, calendar view
Voice
Voice that compounds: capture sessions and learned style
Voice in Contentelli is captured, not prompted. This release wires the full ladder: opinion capture sessions extract stances, cadence, and signature moves into a personal voice model, and once enough approved pieces exist, a learned style fingerprint blends in with few-shot retrieval from your own approved copy. The more you approve, the more the drafts sound like you.
Opinion sessions: record or paste raw takes, get a structured voice model
Learned tier: style fingerprint + retrieval from approved copy after five approved pieces
Visible learning: the product shows which voice tier produced each draft and why
Performance
Faster boot, faster drafts
A performance pass on the two moments that matter most: opening the app and waiting for a draft. App boot no longer blocks on role resolution, settings reads are edge-cached, and draft loading paths were cut down so the review queue is ready when you are.
Boot no longer waits on role and voice-tab resolution
Edge cache in front of settings reads
Leaner draft loading on the review and drafts surfaces