Griptape Agent Config: Azure OpenAI Drivers

The Azure OpenAI Drivers config is the modern all-in-one node for running a Griptape agent on Azure's OpenAI service. It bundles three drivers into one CONFIG: a chat prompt driver, a dall-e image-generation driver, and an embedding driver - plus a local vector store on those embeddings. If your team gets models through Azure rather than OpenAI's own API, this is the config you'll reach for.

Azure's model is different from OpenAI's in one way that shapes everything this node does: you don't reference models, you reference deployments. The dropdown labels (gpt-4o, gpt-4, ...) are conveniences; what Azure actually gets is the deployment name you type in. And because this node wires three drivers, you're now juggling three deployment names - one for chat, one for image generation, one for embeddings. Getting those right is the whole game with this node.

How it works

The node constructs an AzureOpenAiDriversConfig from three sub-drivers: AzureOpenAiChatPromptDriver for reasoning, AzureOpenAiImageGenerationDriver (dall-e-3 by default) for image tasks, and AzureOpenAiEmbeddingDriver (text-embedding-3-small by default), with a LocalVectorStoreDriver built on the embedding driver. Chat and image each take their own endpoint and key env vars - the pack even supports separate AZURE_OPENAI_DALL_E_3_ENDPOINT / AZURE_OPENAI_DALL_E_3_API_KEY vars if your image deployment lives on a different resource.

The inputs that matter

• model - chat model label: gpt-4o (default), gpt-4, gpt-3.5-turbo-16k, gpt-3.5-turbo.
• deployment_name - the actual chat deployment in your Azure resource. Default gpt-4o, but it must match your portal's deployment name or calls fail.
• image_generation_model / image_deployment_name - dall-e-3 (default) or dall-e-2, plus that model's deployment name.
• embedding_model - text-embedding-3-small (default), text-embedding-3-large, or text-embedding-ada-002. Embeddings on Azure are also deployment-based, via the chat resource's endpoint.
• endpoint_env_var / api_key_env_var - env var names (defaults AZURE_OPENAI_ENDPOINT / AZURE_OPENAI_API_KEY), plus the separate image endpoint/key vars if needed.
• size - the dall-e output size (up to 1024x1792 / 1792x1024).

Output is a single CONFIG for agent creation.

Install and credentials

Same pack - ComfyUI Manager (search "Griptape") or:

cd ComfyUI/custom_nodes
git clone https://github.com/griptape-ai/ComfyUI-Griptape

Endpoint and key from your Azure OpenAI resource; set AZURE_OPENAI_ENDPOINT and AZURE_OPENAI_API_KEY in Settings → Griptape or your environment. Standard caveats: griptape[all] pulls in torch (cu121 reinstall on Nvidia if ComfyUI breaks).

Common gotchas

The trio of deployment names is where this node bites. Each driver sends its own deployment name to Azure, so you need three names that exist in your resource - and the embedding driver uses the chat endpoint, which can silently fail if your admin hasn't deployed text-embedding-3-small there. Check the Azure portal's Deployments list before debugging anything else. Also note the description in ComfyUI mentions AZURE_OPENAI_ENDPOINT_3 / AZURE_OPENAI_API_KEY_3 - a leftover from an older convention; the actual env var defaults are the non-_3 names. If your graph uses the deprecated Azure config too, don't leave both in the same workflow - two configs fighting over Defaults.drivers_config is a quiet way to get the wrong driver.