What we shipped. For endpoint-level changes, see the API changelog.
Brand voice extraction now operates on up to 8 sample emails (previously 5) and produces a richer descriptor including tone axes, vocabulary do's and don'ts, and structural patterns. The descriptor is automatically injected as cached system context on every rewrite for the workspace — no per-call configuration required.
This is the foundation for team brand voice, shipping later this quarter. Existing workspaces with brand voices configured don't need to re-run extraction; the new descriptors backfill on next use.
The score panel now highlights specific phrases that sound AI-generated rather than just returning a single ai_likeness score. Hover any flagged phrase to see why it tripped the detector and click for one-tap rewrite. The detector targets the patterns Claude itself penalizes — "delve," "leverage," "robust" used as verbs — plus structural tells like aggressive symmetry and over-hedged opens.
This is a Haiku pass on top of a local heuristic; the combination cuts false positives substantially compared to either alone.
Sequence builder now runs an inter-step coherence pass after generating all steps. If step 3 references the offer from step 1 in a way that contradicts step 2, or if the tone drifts mid-sequence, the coherence checker flags it before the user persists. The check is a single Sonnet pass over the full sequence; output is line-level annotations the user can resolve or override.
This shipped with no UI change to the sequence builder; the coherence panel appears automatically after generation completes.
Localization is live for ES, FR, DE, and PT-BR. The prompt is explicitly tuned for cultural adaptation, not literal translation — German output uses formal address by default where US English would use first names, French output preserves the verbal-flourish patterns that translate poorly word-for-word, etc.
Translation that ignores culture is worse than no translation at all in most B2B contexts; we did the work to ship something that doesn't read like a robot. Try it on any rewrite by selecting target language in the variants panel.
Render your HTML email as it appears in Gmail web, Outlook desktop, iOS Mail, and dark mode — generated server-side via Playwright. Renders attach to the email record and stay there for version comparisons.
Beta caveat: Outlook desktop rendering still has occasional CSS quirks compared to the actual desktop client. We're working on tighter parity. Reply via in-app feedback if you spot a discrepancy that matters for your use case.
The public API (rewrite, score, subject-lines, preheader, CTA, sequences, roast, teardown, A/B predict, localize) is in private preview. Pro and Team plans can request access via in-app settings.
The API uses bearer tokens minted at workspace level with scoped permissions. Rate limits are per-workspace, not per-key. Full documentation at /api/docs. Webhook delivery for async operations (sequence generation, batch scoring) ships in 0.6.x.