Rocket

More than 25 connectors wire outside services into the apps Rocket builds from a prompt: Stripe and Razorpay for payments, Supabase for the database, sign in, storage and edge functions, Resend, SendGrid, Brevo, Mailchimp and Twilio for email and messages, HubSpot for CRM, Airtable, Strapi, Directus and Webflow for data and content, Google Analytics, Mixpanel, Typeform, Tally and Calendly, and OpenAI, Anthropic, Gemini, Perplexity and ElevenLabs for the app's own AI features.

Any REST API can be imported from Postman, cURL or an OpenAPI spec, and Rocket writes the connection code, error handling and data mapping and binds the endpoint to a chosen screen element. Keys are kept as encrypted environment variables, and the agent pushes code to GitHub and deploys to Netlify.

Sourcedocs.rocket.new/build/connectors/overviewread 2026-10-18

Each Build task plans the app's architecture, writes code across screens, navigation, logic and integrations, and fixes errors it detects on its own before returning a live preview.

When the coding agent fails at a fix twice, faces an architectural choice, starts a large refactor or finishes complex code, it calls the Advisor, a read only subagent on Claude Opus that returns root causes, numbered steps and a table of trade offs for the coding agent to carry out.

Solve breaks a business question into chains of research across live sources and returns a structured report with a recommendation, and long Build and Solve tasks keep running in the background after the tab is closed.

Sourcedocs.rocket.new/build/editor/advisor-agentread 2026-10-18

Launchpad grounds a build or a research task in a team's existing material: paste a link or pick a connector and Rocket reads Notion pages, Google Docs and Sheets, Linear issues, Airtable bases, a Supabase schema, Mixpanel events or Directus collections, extracting features, screens, workflows and data models before it generates anything.

Confluence wiki pages and architecture docs can serve as generation context too. Project files take Markdown, PDF, Word, PowerPoint, Excel, CSV, HTML and images and stay available to every task in the project, while Solve cites the exact records and sources behind its findings.

Sourcedocs.rocket.new/getting-started/task/launchpadread 2026-10-18

Any saved version can be rolled back, labeled or launched, and web apps go to a private staging URL first, reaching production only when a person picks the version and sends it live. For Next.js TypeScript projects, code pushed to GitHub arrives as a pull request to main for review, and support staff can change a project only after the owner grants temporary access, which can be revoked from project settings. The agent writes code and connects services as soon as a message is sent, with no plan to approve first, and a rollback discards every later version for good.

Sourcedocs.rocket.new/build/editor/versionsread 2026-10-18

Rocket reports SOC 2 and ISO 27001 compliance alongside GDPR and CCPA. Access runs at three levels, workspace, project and task, with Owner, Editor and Viewer roles, invitations can include or exclude existing work, and owners set a lifetime credit limit for each Editor from 0 to 10,000. Private Build tasks on paid plans stay even from people holding the link, paid plan projects are never used for AI training while free plan builds are always public and may be, connector keys are encrypted at rest, and single sign on with SAML comes through sales.

Sourcerocket.newread 2026-10-18

Each Build reply shows a version number with the files it changed, and a side by side diff opens every edit with added and removed lines marked, while progress messages name the file being written as generation runs. Long tasks report completion through browser notifications, and Solve reports carry the evidence and a source quality rating behind each finding. Task history stays readable in chat, but a step by step record of the agent's commands and tool calls, or an export of agent activity to outside monitoring, is not offered.

Sourcedocs.rocket.new/build/editor/versionsread 2026-10-18

Inside a project, every task can draw on the same uploaded files, connected Google Drive and Notion documents and the outputs of earlier Solve and Build tasks, so research findings shape what gets built and branding carries into a companion landing page.

Typing @ brings the conversation, findings and conclusions of a chosen earlier task into the current one, and each Build task keeps its full chat and every saved version.

Context stays inside one project, and there is no memory of preferences that follows a person across projects or that they can view and correct apart from the files and tasks themselves.

Sourcedocs.rocket.new/getting-started/project/context/overviewread 2026-10-18

Rocket runs as a hosted service on the web and in an iOS app. Web apps it builds are hosted on Rocket's Vercel account or deployed to the customer's own Netlify account, with separate staging and production environments and custom domains, and mobile apps ship as APK downloads or to Google Play and the Apple App Store. Paid plans can download the full source as a zip or sync it to GitHub and run it anywhere. Data localization is offered through sales without published regions, and the builder itself cannot be self hosted.

Sourcedocs.rocket.new/build/launch-web/launch-your-siteread 2026-10-18

More than 25,000 templates cover websites, landing pages, dashboards, portfolios and online stores across 15 industry categories, from technology and professional services to healthcare, restaurants and real estate, and a build can start from any of them. Prompt starters are grouped for founders, product managers, designers, developers and marketers, recipes walk through a SaaS app, an online store, an AI app, a booking app or an internal tool, and workflows run from research to launch or a competitive response. Remix links let anyone copy a shared project and continue it.

Sourcerocket.new/templatesread 2026-10-18

Work begins with a prompt in the web app or the iOS app, with screenshots, PDFs, spreadsheets or documents attached, a pasted Notion, Linear or Google link, a URL to redesign, a template, or a GitHub repository to import.

Long Build and Solve tasks finish in the background and send a browser notification when done, and the agent sleeps after 10 to 15 minutes without activity until someone wakes it.

Intelligence, which watched competitors continuously and delivered daily, weekly or monthly briefs, is paused for new sessions, and nothing else starts the agent on a schedule, an event or a message in another channel.

Sourcedocs.rocket.new/getting-started/task/overviewread 2026-10-18

Rocket picks the models, and the builder has no model setting: initial generation and follow up edits run on Anthropic's Claude Sonnet 4.6, and the Advisor subagent runs on Claude Opus for the hardest moments. Solve routes each question to a light or full research mode on higher plans, which sets depth rather than provider. The OpenAI, Anthropic, Gemini, Perplexity and ElevenLabs connectors power AI features inside the app being built, on the customer's own keys, and do not change which model builds it.

Sourcedocs.rocket.new/changelog/2026/february/week-4read 2026-10-18

Outside tools cannot drive Rocket: there is no API, SDK, command line tool or MCP server for creating tasks, prompting the agent or publishing an app. Code leaves through a zip download on paid plans and through GitHub sync, which is two way for Next.js TypeScript projects so edits made in a local IDE or by teammates can be pulled back in, and push only for Flutter and other frameworks. Remix links copy a project into another account, and custom code injection adds scripts and snippets to a generated app.

Sourcedocs.rocket.new/build/connectors/github/code-syncread 2026-10-18

When Rocket detects an error after a change it fixes it automatically, and the Fix it button on detected errors costs no credits on paid plans. The Advisor checks complex generated code for correctness, edge cases and failure modes before it is finalized, accessibility audits test every page against WCAG 2.1 AA and correct issues in chat, and the Performance view grades website projects from A to F on Core Web Vitals with what to fix. Testing otherwise means clicking through the preview at desktop, tablet and phone sizes or on staging; no test suite is written or run against the app.

Sourcedocs.rocket.new/build/editor/advisor-agentread 2026-10-18

Redesign commands take the URL of a live website, analyze the page and rebuild it with a new look, a mobile first layout or a brand matched design, keeping or rewriting its content, so the agent reads public web pages as source material. The preview is a browser inside the editor that a person clicks through, with screenshots that can be dragged into chat. No agent opens a browser on its own to click through a site, fill in forms or test the app, and none controls a desktop or other software.

Sourcedocs.rocket.new/build/create/commands/redesignread 2026-10-18