RepoReplicateReplicatepublished Oct 29, 2025seen 5d

replicate/cloudflare-ai-gateway-replicate-test

Shell

Open original ↗

Captured source

source ↗

replicate/cloudflare-ai-gateway-replicate-test

Language: Shell

Stars: 0

Forks: 0

Open issues: 0

Created: 2025-10-29T16:55:14Z

Pushed: 2025-10-29T17:12:20Z

Default branch: main

Fork: no

Archived: no

README:

Cloudflare AI Gateway with Replicate

This repository contains a bash script to test Cloudflare AI Gateway with Replicate using the anthropic/claude-4.5-haiku model.

See the official Cloudflare documentation for more information about using Replicate with AI Gateway.

Setup

1. Create a Cloudflare AI Gateway by following the getting started guide

2. Copy .env.example to .env:

cp .env.example .env

3. Fill in your credentials in .env:

  • CF_ACCOUNT_ID: Your Cloudflare account ID (find it here)
  • REPLICATE_API_TOKEN: Your Replicate API token from https://replicate.com/account/api-tokens

Usage

Run the script with an optional prompt:

./replicate-test.sh "Your prompt here"

Or use the default prompt:

./replicate-test.sh

Parameters

Prompt (optional): The prompt to send to Claude. Default: "Write a haiku about Cloudflare"

The script uses a 30-second timeout with the Prefer: wait header, which tells Replicate to keep the connection open and wait for the model to complete, rather than returning immediately with a prediction ID.

How it works

The script: 1. Loads environment variables from .env 2. Validates required credentials are present 3. Constructs the Cloudflare AI Gateway endpoint for the ai-gateway-ziki gateway 4. Makes a POST request to create a prediction with the Claude 4.5 Haiku model 5. Uses standard Replicate authentication with Authorization: Bearer header 6. Includes Prefer: wait header to get synchronous responses 7. Routes the request through Cloudflare AI Gateway for monitoring and caching

Endpoint Format

https://gateway.ai.cloudflare.com/v1/{account_id}/ai-gateway-ziki/replicate/predictions

Example Response

The script will output the JSON response from Replicate, which includes the model's completion when using Prefer: wait.

Notability

notability 1.0/10

Trivial test repo, no traction