Repository navigation
Description
Android App v2026.5.25 node mode chat.send never reaches Gateway v2026.6.5
Environment
- Gateway: 2026.6.5 (5181e4f)
- Android App: 2026.5.25
- Device: Xiaomi 17 Ultra
- Network: Tailscale wss://
- Pairing: Setup Code + autoApproveCidrs
Symptoms
- Pairing succeeds, device shows node + operator roles
- Chat panel can pull history (chat.history OK)
- WebChat replies are pushed to phone correctly (WebSocket push OK)
- When tapping send on phone: input clears, message appears in chat, no error shown
- BUT chat.send never appears in Gateway logs
Verification (Gateway side)
openclaw devices list shows correct scopes:
Xiaomi 17 Ultra | Roles: node, operator | Scopes: operator.read, operator.write, operator.approvals, talk.secrets
Gateway logs - operator connection reads history correctly:
14:46:29 res OK chat.history 548ms conn=9f479141
14:46:31 res OK chat.history 184ms conn=9f479141
WebChat sends work normally (for comparison):
res OK chat.send runId=... conn=f6a68007
chat.send from Android never logged after tapping send.
References
- https://docs.openclaw.ai/gateway/operator-scopes
- https://docs.openclaw.ai/cli/devices
- https://openclaw.cc/platforms/android.html
Analysis
chat.send is missing only in the phone-to-gateway direction. The operator connection works for reading history and receiving push. Appears to be a client-side issue where the Android app's send button does not actually fire chat.send over the operator WebSocket, or is intercepted by local session state before sending. Consistent with SendButton/healthOk/pendingRunCount analysis from community.
Metadata
Metadata
Assignees
Labels
Type
Fields
Priority
Activity
clawsweeper commentedon Jun 10, 2026
Codex review: keeping this open for maintainer follow-up; there is still a little grit to resolve. Reviewed June 21, 2026, 4:43 PM ET / 20:43 UTC.
Summary
Keep open. Current main and v2026.6.9 still show Android chat send wiring and status-blind optimistic UI handling, but neither source nor the open related PRs proves the reported phone-to-Gateway
chat.sendframe loss is fixed.Reproducibility: no. The reporter supplied concrete Gateway-side logs and current source shows the Android UI routes to
chat.send, but this review did not reproduce the Android/Tailscale/Gateway missing-frame path on current main.Ways to help us reproduce this
Next step
The issue already has no-new-fix-pr, needs-maintainer-review, and needs-info labels, and the missing-frame symptom needs live Android proof before an automated repair can choose the correct code path.
Review details
Best possible solution:
Run a maintainer-owned Android live reproduction against current Gateway with Android logs and Gateway request logs; if the frame is still absent, fix the Android outbound RPC/session layer or compose it with the ACK/composer work already open.
Do we have a high-confidence way to reproduce the issue?
No. The reporter supplied concrete Gateway-side logs and current source shows the Android UI routes to
chat.send, but this review did not reproduce the Android/Tailscale/Gateway missing-frame path on current main.Is this the best way to solve the issue?
Not yet. The best fix cannot be chosen until live proof distinguishes a pre-send Android gate, WebSocket transport failure, Gateway auth/scope rejection, or terminal ACK/status handling issue.
AGENTS.md: found and applied where relevant.
Remaining risk / open question:
chat.sendframe remains unverified on current main.Codex review notes: model internal, reasoning high; reviewed against 2609b9722280.
Label changes
Label justifications:
P1: The report describes a user-facing Android chat workflow where outbound messages appear locally but are not delivered to Gateway.impact:message-loss: The central symptom is a phone-originated chat message being lost before Gateway receives or logschat.send.Evidence reviewed
Acceptance criteria:
chat.sendrequest.What I checked:
AGENTS.md:1, 2609b9722280)chat.send; current labels still request maintainer review and more info.chat.sendparams, callssession.request("chat.send", ...), then treats a runId-like response as accepted without checking terminal ACK status. (apps/android/app/src/main/java/ai/openclaw/app/chat/ChatController.kt:246, 2609b9722280)apps/android/app/src/main/java/ai/openclaw/app/ui/chat/ChatScreen.kt:221, 2609b9722280)reqframe and sends it via OkHttp WebSocket; a socket send failure throws locally, so a live trace is needed to distinguish pre-send gating from transport or Gateway receipt. (apps/android/app/src/main/java/ai/openclaw/app/gateway/GatewaySession.kt:360, 2609b9722280)chat.sendis anoperator.writeGateway method, and the Android docs describe Android Chat send aschat.send, matching the reporter's expected contract. (src/gateway/methods/core-descriptors.ts:211, 2609b9722280)Likely related people:
apps/android/app/src/main/java/ai/openclaw/app/chat/ChatController.kt,apps/android/app/src/main/java/ai/openclaw/app/ui/chat/ChatSheetContent.kt,apps/android/app/src/main/java/ai/openclaw/app/gateway/GatewaySession.kt)apps/android/app/src/main/java/ai/openclaw/app/chat/ChatController.kt,apps/android/app/src/main/java/ai/openclaw/app/ui/chat/ChatScreen.kt,apps/android/app/src/main/java/ai/openclaw/app/ui/chat/ChatSheetContent.kt)apps/android/app/src/main/java/ai/openclaw/app/gateway/GatewaySession.kt,apps/android/app/src/main/java/ai/openclaw/app/NodeRuntime.kt)apps/android/app/src/main/java/ai/openclaw/app/chat/ChatController.kt,apps/android/app/src/main/java/ai/openclaw/app/NodeRuntime.kt)How this review workflow works
@clawsweeper re-reviewcomments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.@clawsweeper re-reviewor@clawsweeper re-runon an open PR or issue to request a fresh review only.@clawsweeper reviewto request a fresh review only.@clawsweeper autofix,@clawsweeper automerge,@clawsweeper fix ci, or@clawsweeper address review.@clawsweeper explainto ask for more context, or@clawsweeper stopto stop active automation.elisspace commentedon Jun 30, 2026
I’m seeing the same symptom on a newer setup.
Gateway: OpenClaw 2026.6.10 (aa69b12)
Android app/node reports: 2026.6.9
Device: Pixel 9 Pro, approved/connected, roles node + operator
Operator scopes:
operator.approvals,operator.read,operator.talk.secrets,operator.writeAndroid Main chat can display history. A server-side diagnostic assistant reply written into the same mobile session appeared in the Android app after refresh, so
chat.history/ display works.But when sending from Android, the user message is appended to the session transcript and no assistant response starts. During send attempts, gateway logs show no
chat.send, nosessions.send, and no agent run. Session token usage does not change. One captured Android send appended a user row at2026-06-30T15:08:56Zwith no matchingchat.sendor agent run around that time.The phone lacks
operator.pairing, sodevice.pair.listlogsmissing scope: operator.pairing, butchat.historyrequiresoperator.readandchat.sendrequiresoperator.write, both present. This seems consistent with the Android client persisting/syncing the user row without invokingchat.send, rather than a gateway/model/auth problem.secondfoxgit commentedon Jul 2, 2026
I can reproduce the same issue on a newer setup.
Environment
2026.6.112026.6.9/ build2026060901wss://gateway endpointnode+operatorrolesObserved behavior
The Android app connects successfully and can display chat history. Android node transport also works: gateway-side
device.statusanddevice.infocalls to the phone succeed.However, sending from the Android Chat tab does not start an assistant run:
sessions.listandchat.history.chat.sendfor the Android-origin send attempt.I tested both the default
Mainlane and another selected session lane. In both cases the user message was stored, but nochat.sendreached the gateway.This does not look like a gateway reachability, model/provider, Tailscale, or node approval issue. It looks like the Android Chat client stores/syncs the user message without invoking
chat.send.steipete commentedon Jul 2, 2026
Maintainer live proof: closing this as fixed by #91049 /
3ff0c29f9d09cbcbadc7b2c6855bb73bb94364c0, included inv2026.6.11.I reproduced the reported matrix on a clean Android 16 emulator: Android APK
v2026.6.9-> Tailscale Servewss://-> Gatewayv2026.6.5. The uniquely marked Android send was written to the isolated Gateway session transcript. I then deliberately removed downstream provider auth: thev2026.6.9client showed the reported silent user row, while the same send from thev2026.6.11Android tree displayed the exact Gateway agent failure. Finally, current Android -> current Gateway over the same WSS path completed a real model turn and rendered the reply.The missing
chat.sendconsole line was a misleading signal, not evidence that the frame was absent. Normal Gateway WebSocket logging records incoming request timing but, outside verbose mode, only prints failed or slow responses; a fast acceptedchat.sendcan therefore produce no method line. The persisted user transcript row is stronger evidence: Android sends throughChatController -> GatewaySession.request("chat.send"), and the Gatewaychat.sendhandler owns that transcript write.#91049 fixed the actual user-visible failure mode after the
v2026.6.9APK: terminal send failures and failed agent starts no longer look like a successful silent send. The fix shipped in thev2026.6.11tag. If this still reproduces from an Android build containing that commit, please open a fresh issue with Android logcat plus the correlated Gateway diagnostic/lifecycle error; absence of a normalchat.sendconsole line alone is not a failed-ingress signal.Proof references:
elisspace commentedon Jul 8, 2026
I am seeing what looks like the same failure mode on newer versions, with one extra detail that may help narrow it down.
Environment
2026.6.11(e085fa1)v2026.6.11wss://gateway.REDACTED.compaired · connected · approvednode,operatoroperator.approvals,operator.read,operator.talk.secrets,operator.writeSymptoms
This matches the original report very closely:
chat.sendrequest from Android for the sent message.Additional detail from this repro
In this repro, the Android-originated user message is not only visible locally on the phone: it is also written into the server-side session transcript JSONL.
Example transcript entry after sending
test dispatch 2039from Android Main:{ "type": "message", "timestamp": "2026-07-08T20:42:41.926Z", "message": { "role": "user", "content": "test dispatch 2039", "timestamp": 1783543361743, "idempotencyKey": "9fb163e9-e1eb-454c-9ada-f961a6a52f7:user" } }But no agent run starts afterward. The session age updates, and the user turn is present in the transcript, but dispatch does not happen.
That suggests this may not be only a local optimistic UI issue. The user turn is reaching some persistence/projection path, but it appears to bypass the normal
chat.sendpath that starts the agent turn.Gateway log evidence
Around the Android send attempt, the gateway logs show normal Android connection activity such as:
But there is no corresponding:
for the Android-originated message.
The missing
operator.pairingerror appears unrelated to the chat failure. The Android bootstrap/operator token intentionally excludesoperator.pairing, and the phone is otherwise connected, approved, able to read history, and able to receive displayed replies.Manual dispatch comparison
If I manually run an agent turn against the same node session from the gateway CLI, the assistant response is written to the same session and appears on the phone correctly.
That confirms:
Expected behavior
When Android Main sends a chat message, the gateway should receive/log
chat.send, start an agent run, and stream or deliver the assistant reply back to the same session.Actual behavior
The Android-originated user message is appended to the session transcript, but no
chat.sendappears in gateway logs and no agent run starts.Why I think this is the same issue
The core symptoms match this issue exactly:
chat.sendloggedThe main difference is that this is on
2026.6.11, not2026.5.25/2026.6.5, and the transcript append providesa slightly more specific failure shape: the user message is reaching persistent session state, but not the agent dispatch path.