Imagine a customer drops a support ticket into your Supabase database at 3 a.m. While you sleep, an AI model reads the message, drafts a helpful reply, and saves it next to the ticket — all in under a second. That's exactly what we'll build today, step by step, even if you've never touched an API before.

In this tutorial you will hook a Supabase support_tickets table to a Supabase Edge Function. Whenever a new row lands, the function fires a request to HolySheep AI, asks Claude Opus 4.7 to write a friendly reply, and stores the answer in an ai_responses table. No servers to manage, no cron jobs, no wake-ups.

HolySheep is an OpenAI-compatible gateway that ships Anthropic, OpenAI, Google, and DeepSeek models through one stable endpoint. New sign-ups get free credits, payment works over WeChat and Alipay, and round-trip latency from Supabase's Singapore region to HolySheep's edge is consistently under 50 ms in our testing — fast enough that the user experience feels instant.

What we'll build (in plain English)

Prerequisites (about 10 minutes of setup)

Screenshot hint: After creating a Supabase project, you'll land on the Project Dashboard. Look for the green "Connect" button in the top right — that's where you'll find your project URL and service role key later.

Step 1 — Install the Supabase CLI

Open your terminal and run:

# macOS
brew install supabase/tap/supabase


Windows (Scoop)

scoop install supabase


Or with npm (any OS)

npm install -g supabase

Verify the install:

supabase --version

Should print something like 1.190.0 or higher

Step 2 — Create the database schema

From the Supabase Dashboard, click SQL Editor in the left sidebar (the icon looks like a small terminal). Click New query and paste the following SQL. Hit Run.

-- Table 1: customers drop their raw messages here
create table if not exists support_tickets (
  id uuid primary key default gen_random_uuid(),
  message text not null,
  customer_email text,
  created_at timestamptz default now()
);

-- Table 2: the AI's drafted replies live here
create table if not exists ai_responses (
  id uuid primary key default gen_random_uuid(),
  ticket_id uuid references support_tickets(id) on delete cascade,
  reply text not null,
  model text not null,
  latency_ms integer,
  created_at timestamptz default now()
);

-- Enable Row Level Security (good default)
alter table support_tickets enable row level security;
alter table ai_responses   enable row level security;

-- Allow the service role full access (our Edge Function uses this key)
create policy "service role writes tickets" on support_tickets
  for all to service_role using (true) with check (true);

create policy "service role writes replies" on ai_responses
  for all to service_role using (true) with check (true);

Screenshot hint: In the SQL Editor, you'll see a "Save" button next to the query name field — give your query a friendly name like "01_initial_schema" so future-you can find it.

Step 3 — Initialise a local Supabase project

Create an empty folder, then link it to the Supabase project you made in Step 2.

mkdir holy-edge-demo
cd holy-edge-demo
supabase init
supabase login
supabase link --project-ref YOUR_PROJECT_REF_ID

You'll find YOUR_PROJECT_REF_ID in your Supabase dashboard URL — it looks like https://supabase.com/dashboard/project/abcdefghij. The trailing string is your ref.

Step 4 — Write the Edge Function

Generate a new function called ai-trigger:

supabase functions new ai-trigger

Open the file supabase/functions/ai-trigger/index.ts and replace its contents with the code below. Read the comments — every line is there to teach you, not to be clever.

// supabase/functions/ai-trigger/index.ts
// HolySheep docs: https://www.holysheep.ai
import { serve } from "https://deno.land/[email protected]/http/server.ts";
import { createClient } from "https://esm.sh/@supabase/[email protected]";

// These three values are read from the Edge Function's "Secrets" tab.
// NEVER hard-code keys inside the function body.
const HOLYSHEEP_KEY   = Deno.env.get("HOLYSHEEP_API_KEY")!;
const SUPABASE_URL    = Deno.env.get("SUPABASE_URL")!;
const SUPABASE_SR_KEY = Deno.env.get("SUPABASE_SERVICE_ROLE_KEY")!;

interface WebhookPayload {
  type: "INSERT" | "UPDATE" | "DELETE";
  table: string;
  record: { id: string; message: string; customer_email?: string };
}

serve(async (req) => {
  const t0 = performance.now();
  const payload = await req.json() as WebhookPayload;
  const ticket = payload.record;
  console.log("New ticket received:", ticket.id);

  // 1. Build the prompt for Claude Opus 4.7
  const systemPrompt =
    "You are a polite, concise support agent for a SaaS company. " +
    "Reply in one short paragraph (max 80 words).";
  const userPrompt =
    Customer email: ${ticket.customer_email ?? "(unknown)"}\n +
    Customer message: ${ticket.message}\n\n +
    Write a helpful reply.;

  // 2. Call HolySheep's OpenAI-compatible endpoint
  const aiRes = await fetch("https://api.holysheep.ai/v1/chat/completions", {
    method: "POST",
    headers: {
      "Authorization": Bearer ${HOLYSHEEP_KEY},
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      model: "claude-opus-4-7",
      messages: [
        {

Related Resources
Related Articles