OpenClaw’s Mobile App Is Here, and Local-First AI Is Showing Its Limits

OpenClaw, the open-source AI assistant that pitches itself as the privacy-first alternative to hosted chatbots, finally has native apps for both iOS and Android. The official launch landed on X at 1:41 a.m. UTC on June 30, with the project’s account posting “OpenClaw is now on iOS + Android” alongside direct links to the App Store and Google Play. The post pulled in 10,63,410 views, 5,215 likes, 705 retweets, and 291 replies within hours.

But the debut has not gone to plan. Per Android Authority, multiple users replying to the launch post say the interface feels unfinished, that they could not pair the phone to a Gateway at all, and that some who managed to connect found that “nothing works.” The Google Play listing for the Android app carries a 3.4-star average across 17 reviews, with the earliest user ratings from April already flagging trust concerns and setup failures. The reviews that landed on launch day echo the ones posted in April, which tells a different story than the homepage’s marketing line.

What Launched on June 30

The June 30 announcement came from the @openclaw account, which has been the project’s main public voice since the codebase went public on the OpenClaw source on GitHub. The two companion apps shipped to the same day, a first for the project after months of separate desktop betas. The launch lands at 1:41 a.m. UTC, attached to a tweet carrying the lobster emoji OpenClaw uses as its mascot.

The Android app ships as package ai.openclaw.app and is published under the developer name “Valentina Alexander” with bunsthedev@gmail.com as the public contact. The iOS app sits on the App Store under id6780396132, with the listing naming the same private-developer identity. Google Play reports more than 10,000 installs and a 3.4-star average across 17 user reviews. The launch tweet itself accumulated 10,63,410 views, 5,215 likes, 705 retweets, and 291 replies within hours, a striking engagement number against a small installed base.

The Promise That Doubles as the Constraint

OpenClaw’s self-hostable homepage and install flow leads with the brand line “The AI that actually does things” and lists clearing inboxes, sending emails, managing calendars, and checking in for flights as headline tasks. The project is open-source and runs on the user’s own hardware, installable with a single npm i -g openclaw command or via a curl-based installer. Companion apps for macOS and Windows have already shipped, and the iOS and Android versions are positioned as the mobile counterpart to the same stack. The pitch is an assistant that lives on the user’s own server, never shares data with third parties, and routes every action through the user’s own Gateway.

The asterisk arrived on phones on June 30. The Play Store listing makes it explicit: the Android app is “a companion node, not a standalone gateway,” and it “requires a running OpenClaw gateway on macOS, Linux, or Windows (WSL2).” In other words, the phone is a remote control, and the intelligence still lives on a server the user has to install, configure, and keep running.

The “local-first” architecture is the brand, not a feature bolted on. The same architectural choice that lets the project promise “No data shared with third parties” and “No data collected” on its Play Store listing is the reason the iOS and Android apps cannot work as a download-and-go experience. Users who buy into OpenClaw buy into running their own server, and the friction surfacing on launch day is the same friction the architecture always produced.

• Chat with streaming replies, image attachments, and full session history
• Talk Mode for natural voice conversations with ElevenLabs or system TTS
• Live Canvas surface the assistant controls, with dashboards or any HTML/CSS the agent edits
• Per-permission device access: camera, photos, screen capture, location, notifications, contacts, calendar, SMS, motion sensors
• Setup-code or manual pairing to any reachable self-hosted Gateway
• Data safety section declares no data shared with third parties and no data collected

Where Users Say Nothing Works

The complaints cluster around three failure modes, and most of them appeared within hours of the X announcement. Per Android Authority, several users replying to the launch post said the interface felt unfinished. Others said they could not pair the app with their Gateway at all, and some who did manage to connect reported that “nothing works”.

The Play Store reviews tell a similar story, just dated to April. Reviewer Ian Scoles left a one-line review on April 26 saying the app “didn’t respond to commands during set up.” Another reviewer posted on April 27 noting that the Play Store listing showed the app “released on 7 april 2026 and updated at 5 april 2026,” calling the timeline “imposible” [sic]. A third reviewer, posting as Fam Stack, refused to install the app at all on April 14 over missing trust signals.

Fam Stack’s complaint listed the gaps: no external verification, no support website linked from the listing, no contact information beyond a Gmail address, a low install count. The Play Store developer field shows “Valentina Alexander” with bunsthedev@gmail.com and a separate support inbox at obviyus@openclaw.ai. OpenClaw’s official docs and GitHub repo are public, but they are not surfaced from the store page.

The pairing problem is the most concrete. Users say they can scan the setup code, see the Gateway in the app, and still hit a silent failure when sending the first command. The X post sits at over 291 replies, a high reply-to-like ratio that points to engagement from people with questions or complaints rather than cheerleaders. Several respondents said they liked the idea of the app but called the execution poor and both the Android and iOS versions “very raw,” per Android Authority. The launch-day reaction reads less like a hostile review-bombing and more like the first wave of real users running into the same setup wall the docs warn about.

didn’t respond to commands during set up

The Companion Node Architecture

OpenClaw’s own Android companion-node setup runbook explains why the phone app is shaped this way. The platform page calls the app “a companion node app” and says it requires “a running OpenClaw Gateway” before any chat starts. The Gateway is a Node.js process the user runs on a separate machine, and the phone connects to it over mDNS discovery and a WebSocket.

The default Gateway port is 18789, started with openclaw gateway --port 18789 --verbose. For remote access, the docs recommend Tailscale Serve with an HTTPS endpoint, since plain WebSocket over Tailscale IP addresses is explicitly called out as insufficient for first-time pairing. The Android app keeps a foreground notification alive to maintain the WebSocket connection, and it auto-reconnects to the last gateway on launch. None of this works without a Gateway already running on a host the phone can reach.

That is the architecture, and it is the same architecture the GitHub repo has shipped since the project went public. For a developer running a server at home, the trade-off is fine. For someone who downloaded the Android app’s Google Play listing expecting an app that works on its own, it is not.

1. Install the OpenClaw Gateway on macOS, Linux, or Windows (WSL2)
2. Start the Gateway on port 18789, listening on the local network or via Tailscale Serve
3. Install the Android app from Google Play and open the Connect tab
4. Pair the phone to the Gateway with a setup code or manual host/port
5. Approve the pairing request from the Gateway CLI: openclaw devices approve <requestId>
6. Begin chatting, with Chat history, Talk Mode, and Canvas available once paired

• 10K+ Play Store downloads as of June 30, 2026
• 3.4 stars across 17 Google Play reviews
• 10,63,410 views on the launch-day X post
• 5,215 likes and 705 retweets on the same post
• 291 replies, a reply density that tracks active complaints

What ‘Local-First’ Actually Means Here

OpenClaw’s Play Store data safety section states two things plainly: “No data shared with third parties” and “No data collected.” That is not a marketing claim, since the phone app never sends chat history to OpenClaw’s servers because OpenClaw does not run servers in the model of a hosted chatbot. The Gateway belongs to the user, and the phone is a window into a session the user already controls. Local-first here means owning the entire stack, not just the data.

The trade-off is that every convenience OpenClaw removes from a hosted product, the user has to put back themselves. Pairing requires running openclaw devices approve on the Gateway’s command line after scanning a setup code from the phone. Permissions are granted per-capability on the Android side, but the model catalog, skill plugins, and channel integrations all live on the Gateway. There is no account to create, but there is a server to maintain, and the maintenance is the price of the privacy claim.

The model selection is similarly delegated. OpenClaw is a middleware layer that connects to LLMs the user already pays for, such as Anthropic Claude, OpenAI, or a local model running on the same hardware as the Gateway. The docs describe the assistant as living on the Gateway, with the mobile apps acting as a window into a session that has been running somewhere else. That design is the reason the apps can ship with empty data safety fields and still feel genuinely private to the user paying attention.

Why This Matters Beyond OpenClaw

OpenClaw is the highest-profile open-source AI assistant to attempt a coordinated phone launch in 2026. The bet is that the same people who will not install a hosted chatbot will install a self-hosted one and tolerate a manual setup if the privacy upside is real. The launch tells the rest of the open-source AI world what the ceiling on that bet looks like at the app store gate. A 3.4-star rating and a flood of “nothing works” replies on launch day are not a verdict, but they are the shape of the work that comes next. OpenClaw’s open codebase means the bugs that surface now are the bugs the wider community can chew through.

The fixes that follow will tell the rest of the story. Bug reports on pairing, gateway discovery, and command response are the kind of thing an open-source project can chew through in weeks. The harder question is whether OpenClaw’s pitch can survive a phone-first user base that has never run a Gateway and may never want to.

Frequently Asked Questions

What is OpenClaw?

OpenClaw is an open-source personal AI assistant that runs on a self-hosted Gateway on the user’s own hardware (macOS, Linux, or Windows via WSL2), reachable through messaging apps, a web interface, or the new iOS and Android companion apps.

Why are users complaining about the new OpenClaw mobile apps?

Per Android Authority, users replying to the June 30 launch post on X reported three failure modes: an unfinished interface, an inability to pair the phone to a Gateway, and connected sessions in which “nothing works.” The Play Store reviews left in April show a similar pattern of setup complaints from earlier users.

Do I need a Gateway to use the OpenClaw mobile app?

Yes. The Play Store listing and the project’s docs both state that the Android app is a “companion node, not a standalone gateway” and that it requires a running OpenClaw Gateway on macOS, Linux, or Windows (WSL2) before the phone app will function.

Is OpenClaw really private?

OpenClaw’s Play Store data safety section declares “No data shared with third parties” and “No data collected.” Because the Gateway runs on the user’s own hardware and the phone is a remote client, no chat history is sent to OpenClaw-operated servers.

Where can I download the apps?

The Android app is on Google Play under the package name ai.openclaw.app, and the iOS app is on the App Store under id6780396132; both links are on the project’s June 30 X announcement.