From ca2046b1047574300b98771bc87fb16cc9acd6c7 Mon Sep 17 00:00:00 2001
From: Terry Sutton <saltcod@gmail.com>
Date: Wed, 18 Mar 2026 15:43:34 -0230
Subject: [PATCH] Add edge function embeddings

---
 .../edge-fun-generate-embeddings.md           | 90 +++++++++++++++++++
 1 file changed, 90 insertions(+)
 create mode 100644 skills/supabase/references/edge-fun-generate-embeddings.md

diff --git a/skills/supabase/references/edge-fun-generate-embeddings.md b/skills/supabase/references/edge-fun-generate-embeddings.md
new file mode 100644
index 0000000..ee8bf10
--- /dev/null
+++ b/skills/supabase/references/edge-fun-generate-embeddings.md
@@ -0,0 +1,90 @@
+---
+title: Generate Text Embeddings (Edge Function)
+tags: edge-functions, ai, embeddings, deno
+---
+
+## Generate Text Embeddings (Edge Functions)
+
+Generate high-quality text embeddings using Supabase Edge Functions with the built-in AI inference API (no external embedding provider required).
+
+### Overview
+
+1. Create an inference session once per function instance.
+2. Accept a POST request with JSON `{ "input": "..." }`.
+3. Run `session.run(input)` to produce an embedding.
+
+The `gte-small` (https://huggingface.co/Supabase/gte-small) text embedding model is supported in Supabase's Edge Runtime.
+
+### Common Mistake
+
+**Incorrect:**
+
+```typescript
+// Creating the session on every request (wasteful)
+Deno.serve(async (req) => {
+  const session = new Supabase.ai.Session("gte-small");
+  const { input } = await req.json();
+  const embedding = await session.run(input);
+  return new Response(JSON.stringify({ embedding }));
+});
+```
+
+**Correct:**
+
+```typescript
+// ./supabase/functions/embed/index.ts
+const session = new Supabase.ai.Session("gte-small");
+
+Deno.serve(async (req) => {
+  // Extract input string from JSON body
+  const { input } = await req.json();
+
+  // Generate the embedding from the user input
+  const embedding = await session.run(input, {
+    mean_pool: true,
+    normalize: true,
+  });
+
+  // Return the embedding
+  return new Response(JSON.stringify({ embedding }), {
+    headers: { "Content-Type": "application/json" },
+  });
+});
+```
+
+### Build the Edge Function
+
+1. Initialize Supabase locally:
+
+```shell
+supabase init
+supabase start
+```
+
+2. Create the Edge Function (example name: `embed`):
+
+```shell
+supabase functions new embed
+```
+
+3. Replace `./supabase/functions/embed/index.ts` with the handler above.
+
+### Test it locally
+
+Start a local functions server, then invoke the endpoint:
+
+```shell
+supabase functions serve
+```
+
+```shell
+curl --request POST 'http://localhost:54321/functions/v1/embed' \
+  --header 'Authorization: Bearer ANON_KEY' \
+  --header 'Content-Type: application/json' \
+  --data '{ "input": "hello world" }'
+```
+
+Replace `ANON_KEY` with your project's anonymous key (get it with `supabase status`).
+
+Reference: [Generate Embeddings](https://supabase.com/docs/guides/ai/quickstarts/generate-text-embeddings)
+