cdbx.ai launches browser IDE with AI coding; Pro plans 50% off

cdbx.ai is a browser-based IDE that generates and runs code in response to plain-English descriptions, powered by AI agents woven throughout the workflow. Users open a tab, describe what they want to build, and watch the app materialize without installing packages, configuring environments, or writing boilerplate.

The core experience flows through three steps: Describe (tell cdbx in plain English or pick a template), Build (AI generates all files instantly, and the app runs in the browser immediately), and Ship (one click publishes and generates a shareable live URL). The browser environment includes a full Monaco editor (the same engine VS Code uses), a real terminal with streaming output, an AI pair programmer that reads and writes files, and instant preview—no npm install required.

Beyond chat, cdbx deploys agents in specific roles. The Planning Agent helps users write PRDs, design system architecture, plan research reports, or build slide decks by querying their Notion pages, Linear backlog, GitHub repos, Google Drive, Figma, and Slack on demand—no copy-pasting or tab-switching. Once a plan is done, one click hands off to a new coding experiment with the plan as context. Vibe Coding generates complete apps from plain description. Self-Healing detects and fixes build errors without interrupting the workflow. Visual Loop iterates the UI against the live preview. A Help Assistant answers questions about cdbx inline.

On the infrastructure side, cdbx exposes a Runner API—a single curl endpoint that accepts code (30 languages: Python, Go, Node.js, TypeScript, Java, Kotlin, Rust, C#, Ruby, C, C++, PHP, Swift, Scala, Haskell, Elixir, Clojure, Bash, Perl, Lua, R, Nim, Zig, OCaml, F#, Julia, Dart, Crystal, Groovy, and others) and returns stdout, stderr, and exit code in milliseconds from isolated sandboxes. Each execution runs in a fresh container with network isolation, memory caps, PID limits, and seccomp profiles. The cdbx MCP server exposes run_code as a tool, allowing Claude to call it directly and execute generated code without user intervention.

Supported stacks span React, Vue, Svelte, Angular, Preact, and JavaScript/TypeScript on the frontend; Python, Go, Node.js, Java, Ruby, Bun, Rust, and Kotlin on the backend. Code can be exported as ZIP or pushed to GitHub. Published apps can have custom domains removed of the cdbx badge on paid plans.

Pricing starts Free ($0/month, no credit card, 10 AI credits/month, limited to Haiku + GPT-4o-mini, 100 code executions/month, and a cdbx badge on published apps). The Pro plan costs $20/month (currently 50% off with a 7-day free trial), includes all AI models (Claude Sonnet and Opus, GPT-4o), 100–2,000 AI credits per month, 1-month credit rollover, no cdbx badge, and custom domains. A Team plan adds shared workspace, live collaboration, pooled team credits, member roles, per-member usage analytics, encrypted API keys at rest, and the option to bring your own Anthropic or OpenAI API keys. Credit packs never expire and can be purchased on demand.

cdbx.ai enters a crowded no-code/low-code market by combining a full browser-based IDE (using Monaco, the same editor engine as VS Code) with conversational AI code generation and a modular agent system. Unlike chat-only code assistants, cdbx embeds agents across the entire workflow—planning (PRD and architecture generation), build (Vibe Coding for app generation), error correction (Self-Healing), and UI refinement (Visual Loop)—reducing friction between description and execution.

The platform's integration architecture is noteworthy: instead of preloading context, the Planning Agent and other AI tools query connected services (Notion, GitHub, Linear, Slack, Figma, Google Drive) on demand via MCP (Model Context Protocol) servers. This approach avoids stale data and copy-pasting while remaining compatible with self-hosted or third-party MCP servers. For developers, the Runner API surfaces the same capability—code execution in isolated, hardened sandboxes with network isolation, memory caps, and seccomp profiles—as a curl-callable endpoint, opening it to AI agents running on Claude Desktop or Cursor.

The pricing structure (Free with 10 AI credits, Pro at $20/month with 100–2,000 credits and unlimited model access) and credit rollover suggest a focus on sustained, not one-time, usage. The current 50% Pro discount and 7-day free trial are launch incentives. By lowering barriers to both creation (natural language description) and deployment (one-click publishing), cdbx targets both experienced developers (who gain a serverless sandbox and external-data integration) and non-technical builders (who gain a no-setup environment).