* feat(android): durable offline chat with attachments and history-proof retirement Every chat send is journaled to the per-gateway Room outbox before any network attempt, so process death always has one durable recovery owner. Rows survive gateway ACKs as 'accepted' and retire only once the turn is proven in canonical chat.history by idempotency key; ambiguous outcomes (lost ACK, kill mid-send, gateway restart before the transcript write) park as delivery-unconfirmed for explicit retry, per the fail-closed model #103273 landed. An unproven accepted head briefly holds only its own session's queue, and retrying a parked head re-orders still-queued successors behind it. Offline sends now accept picked images and voice notes; attachment bytes persist as chunked BLOBs (512 KB chunks, 8 MB per message, 48 MB per gateway) admitted and retired atomically with their row. Pre-hello 'main' rows pin to the canonical session at first dispatch so later default-agent changes cannot retarget captured input; slash commands are connection-gated by a persisted epoch (legacy queued command rows migrate to a never-matching sentinel) and never auto-replay across reconnects. The Room store moves to schema v4 on top of v3's data-parking migration, adding the epoch column and attachment tables while preserving queued rows. The queued->sending claim is an atomic compare-and-set shared by the direct dispatch and the flush loop, the direct path waits for the startup recovery sweep before claiming, and failed reconnect attempts republish outbox rows so queued sends stay visible on offline cold starts. Proof: 1151 Android unit tests and :app:ktlintCheck green (new migration, storage, and controller coverage for admission, restarts, claims, pinning, gating, ordering, and byte round-trips); a live emulator scenario on the pre-integration build queued text+image offline, survived force-stop plus a device reboot, resent exactly once on reconnect (single <rowId>:user turn in the gateway transcript), and drained the outbox after canonical-history confirmation. Closes #104087. * fix(android): rearm outbox recovery when direct-send state persistence fails * fix(android): keep direct dispatch alive across caller cancellation The direct send's network phase now runs in the controller scope, so a cancelled UI scope (leaving the chat screen mid-send) can no longer strand a claimed row in 'sending' with no user action available; the dispatch completes and settles the row exactly once. Direct-path state persistence failures also re-arm the startup recovery sweep, mirroring the flush path's fail-closed handling. * fix(android): hand claim-persistence failures to the flush lane A direct-send claim that fails to persist no longer reads as a lost race: the admitted row is handed to the flush lane so a healthy connection still delivers it, and the flush path's fail-closed handling owns any repeated storage failure instead of the UI reporting success with no active owner. * fix(android): enforce connection gating and fail-closed parking on every send path The direct dispatch now rechecks a slash command's connection epoch after its durable claim, so a reconnect between admission and dispatch parks the command instead of replaying it on the new connection. Parking a stale gated row only counts once the write persists; a storage failure drops health, re-arms recovery, and halts the flush pass instead of spinning or dispatching the stale row. * fix(android): recognize gateway-acknowledged run ids in outbox ownership checks A chat.send ACK can return a run id that differs from the row's idempotency key (the direct path transfers local run ownership to it). Ownership, in-flight, backlog, and timeout-park checks now consider both ids, so reconciliation can no longer park — and Retry can no longer duplicate — a turn that is still running on the gateway. * fix(android): close remaining ack-ownership and persistence-failure gaps in the outbox Flushed sends acked under a divergent run id now transfer local run ownership like the direct path, so live runs cannot time out and surface spurious errors for delivered turns. A session pin that cannot persist stops the dispatch while the row is still safely queued. Reconcile and timeout parking only report changes their writes actually persisted, failing closed on storage errors. * docs(android): record why acknowledged run ids stay in-memory
20 KiB
summary, read_when, title
| summary | read_when | title | ||||
|---|---|---|---|---|---|---|
| Android app (node): connection runbook + Connect/Chat/Voice/Canvas command surface |
|
Android app |
Support snapshot
- Role: companion node app (Android does not host the Gateway).
- Gateway required: yes (run it on macOS, Linux, or Windows via WSL2).
- Install: Google Play or
OpenClaw-Android.apkfrom a supported GitHub Release, Getting Started for the Gateway, then Pairing. - Gateway: Runbook + Configuration.
- Protocols: Gateway protocol (nodes + control plane).
System control (launchd/systemd) lives on the Gateway host — see Gateway.
Install outside Google Play
Regular final and correction GitHub Releases include a universal OpenClaw-Android.apk and OpenClaw-Android-SHA256SUMS.txt. The APK is built from the release tag, signed with the OpenClaw Android release key, and carries GitHub Actions provenance.
Choose a release that lists both assets, then download and verify that exact tag before sideloading:
release_tag=vYYYY.M.PATCH
gh release download "$release_tag" \
--repo openclaw/openclaw \
--pattern OpenClaw-Android.apk \
--pattern OpenClaw-Android-SHA256SUMS.txt
sha256sum --check OpenClaw-Android-SHA256SUMS.txt
gh attestation verify OpenClaw-Android.apk \
--repo openclaw/openclaw \
--signer-workflow openclaw/openclaw/.github/workflows/android-release.yml \
--source-ref "refs/tags/${release_tag}" \
--deny-self-hosted-runners
Mirror and control Android from a remote Mac
scrcpy mirrors an Android screen in a macOS window and forwards keyboard and pointer input through Android Debug Bridge (ADB). This is an operator-side workflow, separate from the OpenClaw node connection. It is useful when the Android device and the Mac are in different locations but share a private Tailscale network.
Before you begin
-
Install Tailscale on the Android device and the Mac, and connect both to the same tailnet.
-
On Android, enable Developer options and USB debugging. Android 16 places Wireless debugging under Settings > System > Developer options. See Android developer options.
-
Install scrcpy and ADB on the Mac:
brew install scrcpy brew install --cask android-platform-tools -
Keep the Android device available for the first connection. Android must approve each Mac's ADB key before that Mac can control the device.
Enable ADB over TCP
For the initial setup, connect the Android device by USB to a trusted computer and approve its debugging prompt. Then run:
adb devices
adb tcpip 5555
You can now disconnect USB. If port 5555 stops listening after a device reboot or debugging reset,
repeat this local setup step. Android 11 and later can also establish the initial trust with
Wireless debugging > Pair device with pairing code and adb pair.
Allow only the controller Mac
Tailnets with restrictive grants must explicitly allow the controller Mac to reach TCP port 5555 on the Android device. Add a narrow rule to the tailnet policy, replacing the example addresses with the two devices' stable Tailscale IPs:
{
grants: [
{
src: ["<remote-mac-tailnet-ip>"],
dst: ["<android-tailnet-ip>"],
ip: ["tcp:5555"],
},
],
}
See Tailscale grants for host aliases and other selectors. Do not grant this port to the public internet or expose it with Funnel: an authorized ADB client has broad control of the device.
Connect and start mirroring
On the remote Mac:
adb connect <android-tailnet-ip>:5555
adb devices
scrcpy --serial <android-tailnet-ip>:5555
The first adb connect from this Mac shows an authorization dialog on Android. Unlock the device,
confirm the key fingerprint, and select Always allow from this computer only when the Mac is
trusted. A successful adb devices entry ends in device; unauthorized means the on-device prompt
has not been approved.
Once the scrcpy window opens, use it directly or target it with a macOS screen-automation tool such as Peekaboo. scrcpy carries the display and input; Tailscale provides only the private network path.
Troubleshooting
Connection timed out: verify the tailnet grant for TCP 5555. A successfultailscale pingproves peer reachability, not that policy permits this TCP port. Test withnc -vz <android-tailnet-ip> 5555from the Mac.unauthorized: unlock Android and approve the remote Mac's ADB key, or remove the stale workstation under Wireless debugging > Paired devices and pair it again.Connection refused: reconnect locally and runadb tcpip 5555again.- More than one device listed: keep the explicit
--serial <android-tailnet-ip>:5555argument.
When finished, close scrcpy and disconnect ADB:
adb disconnect <android-tailnet-ip>:5555
Connection runbook
Android node app ⇄ (mDNS/NSD + WebSocket) ⇄ Gateway
Android connects directly to the Gateway WebSocket and uses device pairing (role: node).
For Tailscale or public hosts, Android requires a secure endpoint:
- Preferred: Tailscale Serve / Funnel with
https://<magicdns>/wss://<magicdns> - Also supported: any other
wss://Gateway URL with a real TLS endpoint - Cleartext
ws://remains supported on private LAN addresses /.localhosts, pluslocalhost,127.0.0.1, and the Android emulator bridge (10.0.2.2)
Prerequisites
- Gateway running on another machine (or reachable via SSH).
- Android device/emulator can reach the gateway WebSocket:
- Same LAN with mDNS/NSD, or
- Same Tailscale tailnet using Wide-Area Bonjour / unicast DNS-SD (see below), or
- Manual gateway host/port (fallback)
- Tailnet/public mobile pairing does not use raw tailnet IP
ws://endpoints. Use Tailscale Serve or anotherwss://URL instead. - The
openclawCLI available on the gateway machine (or via SSH), to approve pairing requests.
1. Start the Gateway
openclaw gateway --port 18789 --verbose
Confirm in logs you see something like:
listening on ws://0.0.0.0:18789
For remote Android access over Tailscale, prefer Serve/Funnel instead of a raw tailnet bind:
openclaw gateway --tailscale serve
This gives Android a secure wss:// / https:// endpoint. A plain gateway.bind: "tailnet" setup is not enough for first-time remote Android pairing unless you also terminate TLS separately.
2. Verify discovery (optional)
From the gateway machine:
dns-sd -B _openclaw-gw._tcp local.
More debugging notes: Bonjour.
If you also configured a wide-area discovery domain, compare against:
openclaw gateway discover --json
That shows local. plus the configured wide-area domain in one pass, using the resolved service endpoint instead of TXT-only hints.
Cross-network discovery via unicast DNS-SD
Android NSD/mDNS discovery does not cross networks. If the Android node and the gateway are on different networks but connected via Tailscale, use Wide-Area Bonjour / unicast DNS-SD instead. Discovery alone is not sufficient for tailnet/public Android pairing — the discovered route still needs a secure endpoint (wss:// or Tailscale Serve):
- Set up a DNS-SD zone (example
openclaw.internal.) on the gateway host and publish_openclaw-gw._tcprecords. - Configure Tailscale split DNS for your chosen domain pointing at that DNS server.
Details and example CoreDNS config: Bonjour.
3. Connect from Android
In the Android app:
- The app keeps its gateway connection alive via a foreground service (persistent notification).
- Open the Connect tab.
- Use Setup Code or Manual mode.
- If discovery is blocked, use manual host/port in Advanced controls. For private LAN hosts,
ws://still works. For Tailscale/public hosts, turn on TLS and use awss:/// Tailscale Serve endpoint.
After the first successful pairing, Android auto-reconnects on launch to the active paired gateway (best-effort for discovered gateways, which must be visible on the network).
Multiple gateways
The app keeps a registry of every gateway it has paired with, so you can switch between them without pairing again:
- Settings -> Gateways lists paired gateways with the active one marked. Tap an entry to switch; the app tears down the current sessions and reconnects to the selected gateway.
- The Connect tab shows a quick switcher when more than one gateway is paired.
- Credentials, device tokens, TLS trust, chat history, and queued offline messages are stored per gateway. Switching never mixes state between gateways, and messages queued while offline are delivered only to the gateway they were written for.
- Forget removes a gateway's registry entry together with its credentials, device tokens, TLS pin, and cached chats.
Presence alive beacons
After the authenticated node session connects, and when the app moves to the background while the foreground service is still connected, Android calls node.event with event: "node.presence.alive". The gateway records this as lastSeenAtMs/lastSeenReason on the paired node/device metadata only after the authenticated node device identity is known.
The app counts the beacon as successfully recorded only when the gateway response includes handled: true. Older gateways may acknowledge node.event with { "ok": true }; that response is compatible but does not count as a durable last-seen update.
4. Approve pairing (CLI)
On the gateway machine:
openclaw devices list
openclaw devices approve <requestId>
openclaw devices reject <requestId>
Pairing details: Pairing.
Optional: if the Android node always connects from a tightly controlled subnet, you can opt in to first-time node auto-approval with explicit CIDRs or exact IPs:
{
gateway: {
nodes: {
pairing: {
autoApproveCidrs: ["192.168.1.0/24"],
},
},
},
}
This is disabled by default. It applies only to fresh role: node pairing with no requested scopes. Operator/browser pairing and any role, scope, metadata, or public-key change still require manual approval.
5. Verify the node is connected
openclaw nodes status
openclaw gateway call node.list --params "{}"
6. Chat + history
The Android Chat tab supports session selection (default main, plus other existing sessions):
- History:
chat.history(display-normalized — inline directive tags, plain-text tool-call XML payloads (<tool_call>,<function_call>,<tool_calls>,<function_calls>, and truncated variants), and leaked ASCII/full-width model control tokens are stripped; silent-token assistant rows such as exactNO_REPLY/no_replyare omitted; oversized rows can be replaced with placeholders) - Send:
chat.send - Durable sending: every send (text, picked images, and voice notes) is journaled to a per-gateway on-device outbox before any network attempt, so app termination cannot lose submitted input. Sends queued while offline deliver in order on reconnect with stable idempotency keys, and a send is retired only after the turn is visible in canonical
chat.history— an acknowledgement alone is not treated as proof of delivery. Ambiguous outcomes (lost acknowledgement, app killed mid-send, gateway restart before the transcript write) surface as visible rows with explicit Retry/Delete instead of auto-resending. Slash commands never auto-replay across a reconnect; they park for explicit retry. The queue is bounded (50 messages and 48 MB of attachment bytes per gateway) and unsent rows expire after 48 hours. Composer drafts that were never submitted are not process-durable. - Push updates (best-effort):
chat.subscribe->event:"chat" - Listen: long-press an assistant message and choose Listen to hear it; audio renders via gateway
tts.speakwith the configured TTS provider chain, and on-device system TTS is used when the gateway cannot render audio. Playback stops on session switch, new chat, app backgrounding, or chat close.
7. Canvas + camera
Gateway Canvas Host (recommended for web content)
To have the node show real HTML/CSS/JS that the agent can edit on disk, point the node at the Gateway canvas host.
Nodes load canvas from the Gateway HTTP server (same port as `gateway.port`, default `18789`).- Create
~/.openclaw/workspace/canvas/index.htmlon the gateway host. - Navigate the node to it (LAN):
openclaw nodes invoke --node "<Android Node>" --command canvas.navigate --params '{"url":"http://<gateway-hostname>.local:18789/__openclaw__/canvas/"}'
Tailnet (optional): if both devices are on Tailscale, use a MagicDNS name or tailnet IP instead of .local, e.g. http://<gateway-magicdns>:18789/__openclaw__/canvas/.
This server injects a live-reload client into HTML and reloads on file changes. The Gateway also serves /__openclaw__/a2ui/, but the Android app treats remote A2UI pages as render-only. Action-capable A2UI commands use the bundled app-owned A2UI page.
Canvas commands (foreground only):
canvas.eval,canvas.snapshot,canvas.navigate(use{"url":""}or{"url":"/"}to return to the default scaffold).canvas.snapshotreturns{ format, base64 }(defaultformat="jpeg").- A2UI:
canvas.a2ui.push,canvas.a2ui.reset(canvas.a2ui.pushJSONLlegacy alias). These use the bundled app-owned A2UI page for action-capable rendering.
Camera commands (foreground only; permission-gated): camera.snap (jpg), camera.clip (mp4). See Camera node for parameters and CLI helpers.
8. Voice + expanded Android command surface
- Voice tab: Android has two explicit capture modes. Mic is a manual Voice-tab session that sends each pause as a chat turn and stops when the app leaves the foreground or the user leaves the Voice tab. Talk is continuous Talk Mode and keeps listening until toggled off or the node disconnects.
- Talk Mode promotes the existing foreground service from
connectedDevicetoconnectedDevice|microphonebefore capture starts, then demotes it when Talk Mode stops. The node service declaresFOREGROUND_SERVICE_CONNECTED_DEVICEwithCHANGE_NETWORK_STATE; Android 14+ also requires theFOREGROUND_SERVICE_MICROPHONEdeclaration, theRECORD_AUDIOruntime grant, and the microphone service type at runtime. - By default, Android Talk uses native speech recognition, Gateway chat, and
talk.speakthrough the configured gateway Talk provider. Local system TTS is used only whentalk.speakis unavailable. - Android Talk uses realtime Gateway relay only when
talk.realtime.modeisrealtimeandtalk.realtime.transportisgateway-relay. - Voice wake is implemented in source (
VoiceWakeMode) but the shipping app runtime always forces it tooffon connect — there is no user-facing toggle today. - Additional Android command families (availability depends on device, permissions, and user settings):
device.status,device.info,device.permissions,device.healthdevice.appsonly when Settings > Phone Capabilities > Installed Apps is enabled; it lists launcher-visible apps by default (passincludeNonLaunchablefor the full list).notifications.list,notifications.actions(see Notification forwarding below)photos.latestcontacts.search,contacts.addcalendar.events,calendar.addcallLog.searchsms.searchmotion.activity,motion.pedometer
9. Workspace files (read-only)
The Home overview includes a Files card that browses the active agent's workspace through the read-only agents.workspace.list / agents.workspace.get gateway RPCs: directory drill-down, text and image previews, and export through the Android share sheet. There are no write operations, and previews are size-capped by the gateway.
Assistant entrypoints
Android supports launching OpenClaw from the system assistant trigger (Google Assistant). Holding the home button (or another ACTION_ASSIST trigger) opens the app; saying "Hey Google, ask OpenClaw <prompt>" matches the app's declared App Actions query pattern and hands the prompt into the chat composer without auto-sending it.
This uses Android App Actions (shortcuts.xml capability) declared in the app manifest. No gateway-side configuration is needed — the assistant intent is handled entirely by the Android app.
Notification forwarding
Android can forward device notifications to the gateway as node.event items. This is configured on the device, in the app's Settings sheet — not in gateway/openclaw.json config.
| Setting | Description |
|---|---|
| Forward Notification Events | Master toggle. Off by default; requires Notification Listener Access to be granted first. |
| Package Filter | Allowlist (only listed package IDs forwarded) or Blocklist (default: all packages except listed IDs). OpenClaw's own package is always excluded in Blocklist mode to prevent forwarding loops. |
| Quiet Hours | Local HH:mm start/end window that suppresses forwarding. Disabled by default; defaults to 22:00-07:00 once enabled. |
| Max Events / Minute | Per-device rate limit on forwarded notifications. Default 20. |
| Route Session Key | Optional. Pins forwarded notification events into a specific session instead of the device's default notification route. |
WhatsApp, WhatsApp Business, Telegram, Telegram X, Discord, and Signal notifications are always excluded. Their messages are already owned by native OpenClaw channel sessions; forwarding the Android notification as a separate node event could route a reply through the wrong conversation.