OpenClaw on Android (2026)

v2026.9.1 called out a fuller Android experience. The official app on Google Play is a Gateway companion node: chat, voice, approvals, screen, and device-aware automation. It is not a Gateway you install from F-Droid as a server.

Source of truth for the app: the apps/android README in the OpenClaw repo. Pair with a Gateway using QR, setup code, or manual connection. Credentials stay encrypted on device.

What works

  • Stream replies, pick models and reasoning effort (needs Gateway write access; read-only connections cannot change them).
  • Session permissions in the composer + menu. Native sessions keep a runtime-owned model; Android shows that instead of offering a model change.
  • Context usage ring under Chat actions, including latest tokens and a cost breakdown.
  • Dictation on the microphone; long-press or Talk for voice. Talk setup failures stay in a dismissible Chat dialog.
  • Sidebar folders and session catalogs from the Gateway, including empty folders.
  • Appearance (theme, accent, language) can sync to a writable profile; text size is device-local.

Talk on Android

Official Talk docs: Android realtime uses a Gateway-owned relay when talk.catalog reports the realtime group ready. Android never opens a client-owned WebRTC session. Older Gateways without the capability hint keep native Talk with a visible explanation. Subscription-authenticated Android live audio verification is still called out as pending in those docs—do not assume ChatGPT-login live audio is finished.

iPhone behavior is different (client-owned WebRTC on some OpenAI realtime configs): iPhone Talk.

What it does not replace

  • A public Gateway. Pairing still requires a reachable, authenticated Gateway. Network isolation.
  • Generic child-session forks and new worktrees for native Android sessions (README: unavailable; some message-level forks remain).
  • Editing Gateway folders from the phone beyond what the connected Gateway already supports.

Related