Griptape Embedding Driver: Azure OpenAI

The embedding version of the Azure story: Griptape Embedding Driver: Azure OpenAI produces text embeddings using OpenAI's embedding models - text-embedding-3-small, text-embedding-3-large, text-embedding-ada-002 - but routes them through your Azure OpenAI resource instead of OpenAI's own API.

If your organization's AI budget flows through Azure, this is the driver that keeps embeddings inside that contract, with billing, keys, and data handling all governed by your Azure setup. If you're a solo user with a plain OpenAI key, the plain OpenAI embedding driver is the simpler choice - Azure just adds a deployment layer.

How it works

Same pattern as every driver in this pack: this node is pure configuration. It builds an EMBEDDING_DRIVER object that downstream nodes - vector-store drivers, RAG nodes - use to turn text into vectors. The node itself doesn't call anything until something downstream uses the driver.

The inputs:

• embedding_model - text-embedding-3-small (default), text-embedding-3-large, or text-embedding-ada-002. Small is the cost-effective default; large gives better quality at a premium; ada-002 is the older standby.
• endpoint_env_var - the env-var name for AZURE_OPENAI_ENDPOINT. Your Azure endpoint, referred to by variable name, not typed in directly.
• api_key_env_var - the env-var name for AZURE_OPENAI_API_KEY.

The Azure-specific wrinkle that surprises people: Azure deploys models under deployment names, and the pack derives the deployment from the model you pick. If your deployment is named differently, that's where your setup and the node's expectations can diverge.

Installing

Ships in the ComfyUI Griptape Nodes pack:

• ComfyUI Manager: search "Griptape" → install ComfyUI-Griptape.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/griptape-ai/ComfyUI-Griptape, then restart.

Pack dependencies: griptape[all], openai, python-dotenv, plus git-hosted extensions. The pack-wide torch caveat applies - if Griptape's torch install mismatches ComfyUI's build, reinstall torch with the CUDA index per the README.

Gotchas

The deployment-name mismatch is the classic failure: if the pack's default deployment lookup doesn't match what you actually created in the Azure portal, embedding calls fail even though the key and endpoint are correct - check your deployment names against the model selection. Also, embedding rate limits on Azure are per-deployment and per-minute; big document batches hit them, and the result is throttling errors, not a broken node. And remember the endpoint env var must be the full resource endpoint URL - a truncated or portal-copied-with-trailing-slash value will fail in subtle ways.