Users report Grok Bot deploys projects from a phone

• A phone can reportedly dispatch a full web-project handoff: minchoi's phone report says a user built a website, bought its domain, and deployed through Convex and Cloudflare in two prompts.
• Grok Bot’s work happens on a persistent cloud machine with a browser, filesystem, and terminal, according to aakashgupta's account, so the chat can continue after the initiating device closes.
• The reported project pattern reaches beyond code generation: minchoi's repo example describes a Bot recruiting additional agents for a repository, while minchoi's game-art example says 74 assets were generated, prepared, and wired into a game in about two hours.
• The cloud computer is shared across an account’s roster, a constraint that petergyang's comparison frames as Grok Bot’s built-in persistent-computer advantage and the official FAQ defines as a shared environment.

The official launch post says Bots can be messaged from a phone or desktop and pick up the same conversation on either. In minchoi’s roundup, that interface stretches from a site-and-domain deployment to a Bot that can SSH into a Mac and run local scripts plus AppleScript.

The product unit is an account-wide cloud computer. The official FAQ says Bots share its files, browser sessions, and command-line credentials, while background work continues if the app, laptop, or iPhone closes.

The launch post says Bots can work through websites and tools with no clean API or MCP, then return when an approval is needed. aakashgupta's account describes the reported machine as a VM with a browser, filesystem, and terminal.

minchoi says the reported phone workflow used two prompts to:

1. Build a website.
2. Buy a domain.
3. Deploy it with Convex and Cloudflare.

The post does not identify the site or publish the prompts, but it makes the phone a dispatch surface while the cloud environment performs the build and deployment work.

One entry in minchoi’s repo example calls for a single Bot to become the “boss” of a repository and recruit more agents, also in two prompts. The official launch post documents the adjacent mechanism: Bots can message one another, share context in threads, and coordinate in group chats.

The same roundup says Grok Bot can be SSHed into a Mac to run local scripts and AppleScript. Its security documentation treats that as a separate capability from the cloud computer, with local-command execution set to ask for approval each time by default.

minchoi’s game-art report describes a game-art run that completed these four stages in about two hours:

1. Generate 74 assets.
2. Crop them.
3. Organize them.
4. Wire them into the actual game.

The reported endpoint is an asset pipeline connected to an existing project, not a folder of isolated generations.

petergyang’s tutorial preview lists five named Bot roles:

1. An advisor that creates and manages other Bots.
2. A YouTube researcher for outlier videos.
3. An X scout for viral and funny posts.
4. A digital Marie Kondo for inbox cleanup and subscription savings.
5. A travel concierge.

The same tutorial preview adds a Gamer Bot for installing and playing Doom, Red Alert, and Commander Keen on the cloud machine. It turns the shared computer into a place for both recurring information work and disposable software experiments.

minchoi’s workflow post splits a personal toolchain across six lanes:

• Real-time research: Grok Bot.
• Planning and orchestration: Grok Bot.
• Day-to-day coding and debugging: Grok Build plus Grok 4.6.
• Writing and running tests: Grok Build plus Grok 4.6.
• Complex coding and debugging: GPT-5.6 Sol.
• Frontend: Fable 5.

Only the research and orchestration lanes go to Grok Bot in this example. That places it upstream of code generation, as the agent that gathers context and routes work.

awilkinson reports a Bot called Relationship Coach that receives a day recorded by a Bee Pioneer pin through an API, then runs nightly at 7 p.m. to produce personal and couples-therapy-style feedback. The example adds scheduled interpretation of a personal data stream to the project and repository demonstrations.

The security documentation lists publishing, purchases, deletion, permission changes, and production changes among actions that benefit from explicit approval boundaries. It also says Bots are not separate security boundaries: the account roster shares cloud-computer files, sessions, and credentials, and deleting a Bot does not clear those shared files or browser sessions.