-
Notifications
You must be signed in to change notification settings - Fork 0
Roadmap
What's shipped, what's next, and the long-tail.
The canonical roadmap lives at docs/ROADMAP-V2.md in the repo. This page is a flattened public view.
v0.1.0 – v0.1.8 (April–May 2026)
- Tauri 2 + Node sidecar rewrite of the upstream Exo Electron shell.
- Gmail OAuth + IMAP (
iCloud, Fastmail, Outlook, Yahoo, custom). - Multi-account support with composite
(account_id, *)indexes. - Anthropic + OpenRouter routing.
- Per-feature model selection (analysis / drafts / refinement / calendaring / summary / archive-ready / sender-lookup / agent-drafter / agent-chat).
- Morning Briefing.
- Smart-action key (Space → run agent's recommendation, undo with ⌘Z).
- Awaiting-reply nudges.
- Learned rules.
- Permission tray + audit visibility.
- Calendar V1 UI (day view, suggest-time, create-event-from-thread).
- Extensions V1 (calendar extension; designed to extend).
- Signed + notarized release pipeline (GitHub Actions, fully automated).
- Auto-update via Tauri minisign-signed bundles.
- Free-OpenRouter-model rate limiter.
- DevTools in release builds.
The wow: you open the app in the morning. Inbox is already triaged into "needs you" / "decided" / "delegated" / "killed." Newsletters auto-archived. Drafts waiting in your voice. Tentative meetings on your calendar marked [DRAFT] for one-tap confirmation. 30-second morning briefing in plain English.
What it takes: permission gate (scaffolded), morning briefing panel (V1 shipped), auto-archive rules (learned-rules engine exists), calendar autonomy.
Effort: ~2 weeks of focused work once calendar autonomy lands.
The wow: while walking, "hey Mail, what does Sarah's email need from me?" — read aloud, reply by voice, drafted in your style.
What it takes: macOS shortcut + tray menu, Whisper (local) for STT, conversational loop with the existing tool surface.
Effort: ~1 week.
The wow: mid-thread with a customer, a banner appears: "this is heading toward a refund request. Last 4 customers who sent emails like this churned within 30 days. Suggested next step: jump on a call. I drafted the invite."
What it takes: outcome labeling on thread terminate, embedding index of past thread shapes, fast "does this rhyme with anything bad?" check on new inbound.
Effort: ~2 weeks.
Smart-action already does this for a single thread. Roadmap is to chain it: hit Space repeatedly to walk a 200-email inbox to zero, with high-confidence threads acted on automatically.
Compose opens with 3 drafts the agent thinks you might be writing right now, based on cross-app context: recent calendar events viewed, docs opened in a browser tab (via Chrome MCP), tasks completed in ClickUp/Linear.
Type "the contract Mark sent last week" — semantic search returns it, not a keyword match. Local embeddings of message bodies, vector search via SQLite-vss (or similar).
Select 50 newsletters. Hit Space. Agent runs the right action on each (typically archive). Same gate as single-thread smart-action.
Calendar, ClickUp, Linear, Notion, HubSpot, Slack. The agent can read context from these and act in them via MCP tools.
Tauri supports both; the sidecar is already cross-platform. Mostly a release-pipeline expansion (Linux AppImage, Windows MSI).
A read + approve client. Push notifications for high-priority decisions; tap to approve the agent's draft, archive a thread, schedule.
Adapter on the existing MailProvider interface. Same agent loop, new provider.
The new, better IMAP. Fastmail uses it natively; would simplify some sync paths.
Cross-thread reasoning. "What did we say to Acme about their renewal six weeks ago?" — the agent walks threads, calendar entries, and inbox notes to answer.
Shared accounts (e.g. team@company.com) with role-based agent rules. Out of scope short-term but worth mentioning.
These have been considered and consciously deferred:
- A native iPad app. Mac + iPhone is the priority hardware pair.
- Web app. Native is the differentiator.
- A mail server. AOS Mail is a client; bring your own provider.
- End-to-end encrypted mail (PGP/S/MIME). Niche audience for the complexity it adds.
- Crypto/blockchain anything. Hard no.
The maintainer (matt@mrdula.solutions) sets priorities based on:
- What he'd ship next as a user. First-person dogfooding drives the roadmap.
- Feedback from beta users. If a real user is blocked, that pulls priority forward.
- Where the underlying capability has matured. Voice-to-inbox waits on Whisper's accuracy; cross-app MCP waits on MCP server availability for the target apps.
If you want to influence the roadmap, the most effective path is to use AOS Mail, file specific issues with reproductions, and discuss tradeoffs in the issue thread. Less effective: generic "please add X" issues without a use case.