OpenAI Embeddings Node zeigt keine Embedding-Modelle an / 400-Fehler bei text-embedding-3-small (Self-hosted Docker)

Questions
1

Hi everyone,

I’m struggling with the OpenAI Embeddings node in my self-hosted n8n instance (Docker, latest image, OpenAI Embeddings Node v1.2).

Issue

  • In the Embeddings node, no embedding models (like text-embedding-ada-002 or text-embedding-3-small) appear in the dropdown—only chat/completion models such as GPT-4, GPT-4o, etc.

  • If I manually enter text-embedding-3-small as the model, I get the error:
    400 status code (no body)

  • My OpenAI API key is valid and works with other OpenAI nodes.

  • Pinecone is correctly connected and working.

  • There is no “Use Response API” option in the Embeddings node.

  • I’m running n8n as a Docker container (not via docker-compose).

What I’ve tried

  • Pulled the latest n8n image and restarted the container multiple times (docker pull n8nio/n8n:latest, etc.).

  • Confirmed that the Embeddings node is present in the workflow.

  • Checked if other embedding models are available (none are).

  • Tested my API key directly with OpenAI (it works).

  • Tried various credentials and settings combinations.

Additional Observations / UI Behavior

  • When I first open the Embeddings node, the input on the left side (e.g., from the Markdown node) is displayed correctly.

  • After closing and reopening the node, or after running the workflow, the error occurs and the input on the left side is no longer visible (see attached screenshots).

  • This is confusing, as the input is initially passed through correctly.

Questions

  • Is this behavior known?

  • Are there limitations for embedding models in the Embeddings node for self-hosted/Docker setups?

  • Is there a way to get embedding models to show in the dropdown?

  • Is this a setup issue or a general bug?

  • Is there a workaround to use embeddings in n8n and send them to Pinecone?

2

Looks like this might be related to the OpenAI Embeddings node version or API compatibility. Since you’re using the latest n8n Docker image, try explicitly setting the `OPENAI_API_VERSION` environment variable to `2023-05-15` in your Docker run command. This API version should support the embedding models correctly.

If that doesn’t work, you could also try manually entering `text-embedding-ada-002` as the model name, as it’s widely supported and might bypass the dropdown issue. Let me know if this helps!

3

thank you, achamm - setting the value manually to “text-embedding-ada-002” did the magic.

2 Likes
4

Your welcome! If you’d like, feel free to set one of our replies as the solution! Have a great day!