BUG: embeddings generation error "models/embedding-001 is not found"

[gamal1osama]

Is there an existing issue for this?

• I have searched the existing issues

What happened?

📌 Issue Overview

The get-embedding Edge Function is using an incorrect Gemini API endpoint and model name, causing all embedding generation requests to fail with a "model not found" error. This breaks the AI-powered semantic search and meeting summarization features that depend on embeddings.

🔍 Steps to Reproduce

1. Deploy the current index.ts to Supabase
2. Call the Edge Function with a test request:

   curl -X POST "https://your-project.supabase.co/functions/v1/get-embedding" \
     -H "Authorization: Bearer YOUR_ANON_KEY" \
     -H "Content-Type: application/json" \
     -d '{"text": "Test meeting about AI"}' | jq

3. Observe the error response

🎯 Expected Behavior

The function should successfully generate a 768-dimension embedding vector using the Gemini API and return:

{
  "embedding": [0.123, -0.456, 0.789, ...]
}

🚨 Actual Behavior

The function fails with the following error:

{
  "error": "Error generating embedding: models/embedding-001 is not found for API version v1, or is not supported for embedContent. Call ListModels to see the list of available models and their supported methods."
}

Root Cause:

• Current implementation uses /v1/models/embedding-001 which doesn't exist
• Gemini's embedding model is actually gemini-embedding-001 and requires the /v1beta API version
• Request body includes unnecessary model and taskType fields

📷 Screenshot

Error output:

Verified correct model via Gemini API:

$ curl "https://generativelanguage.googleapis.com/v1beta/models?key=API_KEY" | \
  jq '.models[] | select(.name | contains("embedding"))'

{
  "name": "models/gemini-embedding-001",
  "supportedGenerationMethods": [
    "embedContent",
    "countTextTokens",
    "countTokens",
    "asyncBatchEmbedContent"
  ]
}

💡 Suggested Improvements

Fix Required Changes

Current (broken) code in index.ts:

const embeddingResponse = await fetch(
  "https://generativelanguage.googleapis.com/v1/models/embedding-001:embedContent?key=" + GEMINI_API_KEY,
  {
    method: "POST",
    headers: {
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      model: "embedding-001",  // ❌ Incorrect
      content: {
        parts: [{ text: text }]
      },
      taskType: "RETRIEVAL_QUERY"  // ❌ Unnecessary
    }),
  }
);

Proposed fix:

const embeddingResponse = await fetch(
  "https://generativelanguage.googleapis.com/v1beta/models/gemini-embedding-001:embedContent?key=" + GEMINI_API_KEY,
  {
    method: "POST",
    headers: {
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      content: {
        parts: [{ text: text }]
      }
    }),
  }
);

Changes Summary:

1. ✅ Update API version: /v1 → /v1beta
2. ✅ Update model name: embedding-001 → gemini-embedding-001
3. ✅ Remove model field from request body (redundant - already in URL)
4. ✅ Remove taskType field (not required by v1beta API)

Verification

After applying the fix, the function works correctly:

$ curl -X POST "https://your-project.supabase.co/functions/v1/get-embedding" \
  -H "Authorization: Bearer YOUR_ANON_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "Test meeting about AI"}' | jq

{
  "embedding": [0.034567, -0.012345, 0.056789, ..., 0.023456]  // 768 dimensions
}

---

Impact: This bug prevents all AI-powered features (semantic search, meeting insights, similarity matching) from functioning. Fixing it is critical for the application to work as intended.

Record

• I agree to follow this project's Code of Conduct
• I want to work on this issue