VibeFlare

The problem it solves

Each prototype needs model selection, authentication and usage tracking. VibeFlare puts those pieces behind one endpoint in your Cloudflare account, alongside a browser workspace for everyday tasks. You manage the installation and access; Cloudflare still supplies and bills the underlying services.

Ways to use it

• Give a prototype one API endpoint

  Create a labelled API key and set the app’s base URL to your instance followed by /v1. Choose a model from that instance’s catalog, rather than an unrelated provider’s model name. Revoke the key when the prototype is retired.

• Turn an audio file into a saved transcript

  Open Workspace > Audio, select a compatible model, drop a recording up to 25 MB and click Transcribe. Reopen its saved conversation from History. The selected file is submitted only when you request transcription.

• Explore what embeddings are for

  Embeddings turns text into numerical vectors for similarity search or grouping. Enter short text or import a .txt/.md file, then generate with an embedding model. These vectors are building blocks for your app, not a chatbot reply or a ready-made search database.

Before you start

• A Cloudflare account with the necessary Workers AI access and R2 enabled; review provider limits before creating resources or sending model requests.
• For CLI-managed setup: Node.js 22.12+, Bun and Git. The installer obtains the project-pinned dependencies and Wrangler.
• A browser to authenticate and create the owner account. Keep any existing installation receipt: it identifies the resources that updates may reuse.

Install and set up

1. Choose one installation route

   The README’s Deploy to Cloudflare button is the simplest new install. The following CLI route adds a local ownership receipt for status, updates and safe uninstall. Inspect an existing installation before creating another.

2. Install the CLI

   Clone the repository and run the installer after checking Node and Bun. The ./vf command remains usable from the clone even before adding ~/.local/bin to PATH.

   git clone https://github.com/alexcodeplace/vibeflare.git
   cd vibeflare
   ./install.sh

3. Authenticate and create your installation

   Authenticate in the browser, confirm the Cloudflare account and resource creation, then run setup. It creates the Worker/database/storage resources and records their ownership; it is not a read-only command.

   corepack pnpm --filter @vibeflare/worker exec wrangler login
   ./vf setup

4. Create the owner account and verify health

   Open the printed URL and complete /setup with your passkey or GitHub account. Create an API key to connect vf login to that URL, entering it locally. Use vf status and vf doctor; keep the installation receipt outside the source checkout.

   ./vf status
   ./vf doctor

Use it in your project

1. Choose a task and a compatible model

   Workspace provides Text, Image, Audio and Embeddings tasks. Model choices match the active task. Text/image/audio/embedding conversations can be reopened from History. Image prompt imports accept written .txt/.md prompts, not reference-image editing.

2. Connect an application

   Create a revocable key in API Keys, keep it in the app’s secret configuration and use your instance URL followed by /v1. Exclude paid filters model access; it is not a spending cap.

   https://YOUR-VIBEFLARE-URL/v1

3. Keep updates and cleanup tied to the receipt

   Use vf update from the newer release with the existing receipt to preserve its D1/R2 data. Preview uninstall before considering deletion; the preview command does not delete Cloudflare resources.

   ./vf uninstall --preview

What to know first

Cloudflare resources and AI usage may incur charges. Check the provider’s current prices and limits.