convex/chef

The only AI app builder that knows backend

Chef by Convex – AI‑powered full‑stack app builder

What it is

  • Chef is an open‑source web application that lets you generate full‑stack web apps by talking to a large language model. It runs on top of Convex, an open‑source reactive database, and wires together a backend (database, auth, file storage, background jobs) with a real‑time UI automatically.

Core capabilities

  • Code generation – Sends prompts to Anthropic, OpenAI, Google, or XAI APIs to generate server‑side functions, UI components, and database schemas.
  • Agentic loop – The chef-agent injects system prompts, defines tool‑calling capabilities, and iterates with the model to refine generated code.
  • Built‑in services – Uses Convex’s APIs for authentication, file uploads, real‑time data sync, and background workflows, so the generated app is production‑ready out of the box.
  • CLI (chefshot) – A command‑line interface for interacting with the web UI and managing projects.
  • Test harness – test‑kitchen provides a sandbox to validate the agentic loop.

How it works

  1. Prompt – You describe the app you want in natural language via the web UI.
  2. Agent – chef-agent constructs a system prompt, calls the chosen LLM, and may invoke tools (e.g., file upload, database queries) to refine the answer.
  3. Codegen – The LLM returns TypeScript/React code that is written into the project’s app/ folder and Convex functions in convex/.
  4. Deploy – The generated Convex project can be run locally (pnpm run dev + npx convex dev) or deployed to Convex’s hosted platform.

Repository layout

  • app/ – Front‑end React components, routes, and client‑side sync logic.
  • chef-agent/ – Logic for the LLM interaction loop and tool definitions.
  • chefshot/ – CLI wrapper around the web UI.
  • convex/ – Convex database schema and serverless functions that store chats and user data.
  • template/ – Boilerplate used as the starting point for every generated project.
  • test‑kitchen/ – Automated tests for the agentic workflow.

Getting started

# Clone and install
git clone https://github.com/get-convex/chef.git && cd chef
nvm install && nvm use
npm install -g pnpm && pnpm i

# Configure environment (add Convex OAuth app and LLM API keys)
echo 'VITE_CONVEX_URL=placeholder' >> .env.local
npx convex dev --once   # creates a Convex project

# Run both front‑end and back‑end
pnpm run dev            # starts the Vite dev server
npx convex dev          # starts Convex dev server

Visit http://127.0.0.1:<port>/ and log in with a Convex account to start building apps.

Typical use cases

  • Rapid prototyping of CRUD web apps without writing boilerplate.
  • Teaching developers how to structure a full‑stack Convex project.
  • Experimenting with LLM‑driven code generation pipelines.

Contribution & licensing

  • The project is maintained by the Convex team. Contributions are welcomed via pull requests; see CONTRIBUTING.md.
  • It is released under the MIT license (standard for Convex open‑source projects).

Where to learn more

Related

  • Project
    IvanMurzak/Unity-MCP

    Unity‑MCP is a Unity package that connects LLM agents (Claude, Codex, etc.) to the Unity Editor and to compiled games. It provides 70+ built‑in “tools” (asset management, scene manipulation, scripting, testing, profiling, etc.) that the AI can call, enabling natural‑language creation, debugging, and runtime AI behaviour. Install via a UnityPackage, OpenUPM, or the `unity-mcp-cli`; configure an agent, auto‑generate a skill set, and start issuing commands like “create three cubes in a circle”. Extensible with optional skill packs for Animation, Cinemachine, Navigation, and more. MIT‑licensed.

  • Project
    phodal/auto-dev

    An AI-native multi-agent development platform built on Kotlin Multiplatform that integrates coding, review, and research agents across IDEs and multiple devices.

  • Project
    markusschanta/awesome-jupyter

    A curated directory of tools, libraries, and resources that extend the functionality of Jupyter notebooks for data science, development, and research.

  • Project
    liustack/modlens

    ModLens is a lightweight vision plug‑in for DeepSeek Harness that lets you paste images directly into a chat with a text‑only LLM. It routes the picture to one of several built‑in or reusable vision services (Gemini, OpenAI‑compatible, Anthropic, Antigravity‑CLI, Claude‑CLI, Kimi‑CLI) and returns a structured description for the model to use. Installation is a single command, and the plugin works without extra proxies or config changes.