Share-sheet and email ingestion
Forwarded email and shared content exist as capture types in the schema. The native share extension and an inbound mail address are later work.
What's built, what's next, what's deliberately not
Most product roadmaps are a wish list. This one is closer to a status report — including the things that are missing on purpose.
Shipped
The first release is feature-complete
Every item below works end to end today.
Done
Foundations
Design system, navigation, authentication, database schema with row-level security and its own SQL test suite, and continuous integration running typecheck, lint and tests on every push.
Done
The work engine
Actions, projects, people, statuses and priorities; the Today, Work and action detail screens; completion, due dates and filtering.
Done
Capture
Text, image, file and voice capture with stored originals, capture history and explicit processing states.
Done
AI extraction
The orchestrator, provider abstraction, classification, action, entity and deadline extraction, confidence scoring and the approval UI.
Done
Follow-up engine
Waiting state, follow-up scheduling, the in-app inbox, draft generation and the approval flow, with activity history behind it.
Done
Daily brief and assistant
Structured queries into a compact brief under 250 words; the assistant's read tools, work search and its refusal to answer without evidence.
Done
Meetings and subscriptions
Meeting recording and review with decisions, actions and risks; the subscription service, free-tier limits and the paywall.
Done
Evaluation suite
A fixed dataset of over a hundred cases plus the MVP acceptance tests, including the prompt-injection fixtures.
Next
What stands between here and the App Store
This is operational work, not feature work.
In progress
Go live
A real backend project, a real AI provider key in the server environment, a production bundle identifier, and a first device build.
Next
Server-side transcription and OCR
Voice captures and meetings store real audio, and scans store the real image; the step that reads them automatically is the outstanding piece. Until it lands, you type the gist and Zephyr does the rest — there is deliberately no fake transcription path.
Next
Background notifications
Follow-ups currently surface in the app's own inbox. A push notification while the app is closed needs a dependency and a permissions story that haven't been designed yet.
Next
Store submission
Review, privacy nutrition labels, and finalised legal copy. The privacy policy and terms on this site are marked as drafts until a lawyer has been through them.
Later
Planned, not promised
Persisted assistant conversations
Today each assistant session starts clean. Storing threads is what makes follow-up questions possible.
Semantic search
Search is Postgres full-text today. Vector search would improve both the assistant's evidence lookup and work search.
Team workspaces, properly
The Business plan shares a workspace; per-member briefs and richer delegation are a genuine V2 concern rather than a toggle.
Not planned
Things Zephyr is choosing not to be
These aren't backlog items. They're boundaries.
Autonomous email or calendar changes
Autonomous customer communication
Financial transactions
Slack / Teams / WhatsApp integrations
CRM and EHR integrations
Desktop app
Watch app
Custom model training on your data
Agent marketplaces
Integrations may arrive eventually; the first four rows of that list will not. An assistant that can act on the outside world without approval is a different product with a different risk profile.
Want to be told when it ships?
Join the early access list — it's an email, not a funnel.