Kimosabe.AI

Roadmap

Kimosabe.AI is built in honest phases. We ship Phase 1 as a real product, then expand the surface area without ever weakening the privacy core.


Phase 1 — MVP (current)

Goal: A real, end-to-end encrypted cloud messenger as an installable PWA.

Status Item
✅ Email + password auth via Lovable Cloud
✅ Username discovery (@handle)
✅ Identity keypair (ECDH P-256) generated in browser at signup
✅ Recovery key (one-time, 120-bit) — re-wraps private key
✅ Wrapped private key stored server-side as ciphertext
✅ 1:1 E2EE chat (AES-GCM-256 + ECDH)
✅ Realtime delivery via Supabase Realtime
✅ Disappearing messages — per-conversation TTL (off / 1d / 7d)
✅ Encrypted image attachments
✅ Settings: change password, rotate recovery key, export identity, delete account
✅ Passwordless recovery flow (zero-knowledge edge function)
✅ Landing + /trust page (public threat model)
✅ PWA manifest + 430px mobile polish
✅ pg_cron job to purge expired messages server-side
✅ Delivery + opt-in read receipts
✅ Service worker for offline thread cache + queued sends
🚧 Accessibility audit (WCAG 2.2 AA — see STANDARDS.md)

Phase 2 — Group chats & verified contacts

  • Group chat E2EE via MLS (RFC 9420). No custom ratchet.
  • Asynchronous handshake via X3DH + Double Ratchet for 1:1 forward secrecy & PCS.
  • Safety numbers for out-of-band identity verification (Signal-style).
  • Voice notes (encrypted audio attachments — HPKE envelope, RFC 9180).
  • Sealed sender (v1) for direct messages.
  • Encrypted push notifications — server pushes ciphertext envelope; service worker decrypts.
  • Optional WebAuthn / Passkeys as a second factor.

See STANDARDS.md for the full spec list and REQUIREMENTS.md for the requirement IDs each item satisfies.


Phase 3 — Metadata reduction

  • Sealed sender (v2) for groups.
  • Optional Tor-like relay to hide IPs from the server.
  • Encrypted profile data — display name, avatar wrapped to contacts only.
  • Federated identity attestation so users can prove control of an external handle (Mastodon, GitHub, etc.) without uploading it to us.

Phase 4 — Off-grid (NATIVE app, not web)

The web app cannot do mesh. iOS Safari has no Web Bluetooth, no background execution, and no raw radio access. Phase 4 ships a separate native client that shares the same identity keys and account so users move between web and native seamlessly.

  • React Native or Swift/Kotlin client.
  • Bluetooth + Wi-Fi-Direct mesh for nearby-device messaging when the internet is down.
  • Optional LoRa companion (Meshtastic-compatible) for grid-down scenarios.
  • Web app remains the primary surface; the native app is the "off-grid mode."

Phase 5 — Super-app surfaces

Once the messaging core is rock-solid, expand carefully:

  • Moments-style feed — encrypted to your contact graph, server-blind.
  • P2P payments via a regulated processor (Stripe / Paddle).
  • Mini-app container with a strict permission model — apps cannot read messages.

Phase 6 — AI agents (see docs/AGENTS.md)

Agent frameworks like Hermes, OpenClaw, and Agent Zero plug in as first-class users of Kimosabe — they have their own identity keypair, their own recovery key, and they encrypt to you the same way another human would. No backdoors, no special server access. See docs/AGENTS.md for the full integration contract.


Out of scope (forever)

These are not on the roadmap. They would compromise the core promise.

  • Server-side message search.
  • Plaintext push notifications.
  • Backup of message history to any cloud we control.
  • Government or operator backdoor of any kind.
  • Phone number requirement.
  • Contact-graph upload.