Vibecode

Asking for a feature has the agent wire in the hosted catalog of more than 30 APIs without separate accounts or keys, among them OpenAI, Claude, Gemini, Grok and GLM models, image, video and sound generation, ElevenLabs voice and music, Resend email, Expo push notifications, Apple Maps, weather, film, stock and crypto data and product search, all billed to Vibecode credits. Vibecode Cloud gives the app a database, Better Auth sign in and file storage from one prompt, Supabase or Firebase can be connected instead, and RevenueCat and Stripe add subscriptions and payments with the customer's own keys.

Sourcevibecodeapp.com/docs/integrations/indexread 2026-10-18

Every build runs Claude Code in a cloud sandbox for the project, so the agent plans the work, edits files across the codebase, runs terminal commands, hands subtasks such as polishing the interface to helper agents and checks its own result before it reports back. Native iPhone apps compile on cloud Macs and run in a cloud simulator, and the agent carries a build through signing, upload and App Store submission when asked. A first build usually takes about 10 minutes, and work continues while the builder watches from the web or the iPhone app.

Sourcevibecodeapp.com/docs/getting-started/claude-coderead 2026-10-18

A build can start from a description, from screenshots, from existing code or from a website's address, which the agent rebuilds as a native Swift or React Native app rather than wrapping the original pages in an app shell. The project's own code is the context for every change. There is no knowledge base, document library or retrieval over a team's own material that the agent draws on while it works.

Sourcevibecodeapp.com/docs/getting-started/web-to-mobileread 2026-10-18

A person chooses when to deploy or redeploy a web app, when to install a native build on an iPhone and when to submit to the App Store or TestFlight, and Apple's review stands between a submission and the store. Inside Claude, each Vibecode tool asks for permission once or always before it runs. Within Vibecode the agent changes code as soon as a prompt is sent, with no plan to approve first and no checkpoint or rollback control in the builder.

Sourcevibecodeapp.com/docs/deploying/overviewread 2026-10-18

Builders keep full ownership of the code and data they create, can export and delete projects at any time, and the service is hosted in the United States. Apple Developer access is connected once with an App Store Connect key that Vibecode creates or the builder supplies. There are no team roles, single sign on, admin controls or audit records, and no SOC 2 or other attestation.

Sourcevibecodeapp.com/privacyread 2026-10-18

A Usage tab on the web shows what has been spent by model and by app, with request and token counts. While the agent tests, its browser or simulator appears live in the preview with a testing badge, and inside Claude a Vibecode card shows what the agent is working on during a build. A record of each tool call and command the agent ran, or an export of agent activity to outside monitoring, is not offered.

Sourcevibecodeapp.com/docs/updatesread 2026-10-18

Each project keeps its code, its backend and its chat history, and a new chat can be opened inside a project while earlier chats stay available to return to. Credits roll over each cycle and never expire. The agent keeps no memory of preferences or decisions that carries from one chat or project to the next beyond the code it wrote, and there is nothing a person can view or edit that shapes its behavior across projects.

Sourcevibecodeapp.com/docs/faqsread 2026-10-18

Builds run in Vibecode's cloud sandboxes and on cloud Macs, hosted in the United States, and web apps and backends deploy to a vibecode.run address or a custom domain. Native iPhone apps install on a registered iPhone or go to the App Store and TestFlight, React Native apps reach the App Store through Expo, and Google Play publishing means exporting the Expo project. Pro and Max can download the full Xcode, Expo or web project or connect over SSH; the builder itself cannot be self hosted.

Sourcevibecodeapp.com/docs/deploying/overviewread 2026-10-18

A gallery of working iPhone app templates can be remixed and changed by chatting, from a 3D tower defense game, a stock trading app with candlestick charts and a fitness tracker with activity rings to a meditation app, a fragrance diary, a walking leaderboard, a ride hailing app and clones of familiar social, streaming, dating, payments and travel apps. Starter prompts cover a workout log, a shared budget, a job tracker and simple games.

Sourcevibecodeapp.com/templatesread 2026-10-18

Builds start from a prompt in the web app or the Vibecode iPhone app, from a Claude chat on the web, desktop or iPhone once the Vibecode connector is added, or from another coding agent such as Cursor, Codex or Gemini through the Vibecode CLI. Claude can wait for a build and report when it is ready. Each of these starts with a person or an agent they run; Vibecode itself has no schedules, event triggers or chat channel bots that start work on their own.

Sourcevibecodeapp.com/docs/guides/vibecode-in-clauderead 2026-10-18

A model picker chooses the model for each message from Anthropic's Claude Opus 5.5 and Sonnet 5.5, OpenAI's GPT 6.1 Sol, GPT 6 Luna and GPT 6 Astra, Google Gemini, Meta Muse and GLM 5.3 and GLM 5.3 Flash, all running inside Claude Code. By default a new app's first build uses Claude Opus 5.5 and later changes use GPT 6.1 Sol, and credits bill model use at the providers' own prices with no markup.

Sourcevibecodeapp.com/docs/updatesread 2026-10-18

The Vibecode CLI, installed from its GitHub releases for macOS, Linux and Windows and authorized with an account API key, gives Claude Code, Codex, Gemini CLI, OpenClaw or other agents Vibecode's sandboxes, databases, authentication, APIs and deployments to a custom domain or the App Store, and it prints its own command reference as a skill file for the agent. A remote MCP server at api.vibecodeapp.com/mcp with OAuth sign in lets Claude build and preview Vibecode apps from a chat.

Sourcevibecodeapp.com/docs/vibecode-cliread 2026-10-18

After any change to an app's screens or behavior, the agent opens the app, clicks or taps through what it just built, checks that it does what was asked and fixes what it finds before saying it is done, on every web and mobile project with nothing to set up. Native iOS apps get their own version on the iOS simulator. A builder can name a flow to check, such as signing up with a new account and confirming the welcome screen appears.

Sourcevibecodeapp.com/docs/features/agent-testingread 2026-10-18

The agent drives a real browser, or the iOS simulator for native apps, to click, tap and scroll through the app it just built, and the builder can watch and even click in the same live view. That browser is aimed at the project's own app; the agent does not browse other websites, fill in forms on outside services or operate a desktop. Native iOS checks run on a cloud simulator streamed to the browser or phone.

Sourcevibecodeapp.com/docs/features/agent-testingread 2026-10-18