Griptape Agent Config: OpenAI Drivers

This is the config node most people start with, and honestly, the one you'll probably keep using. If you have an OpenAI key, you drag this in, connect it to an Agent, and you've got an LLM that can chat, describe images, generate images, embed text, talk out loud, and transcribe audio - all from one box. It's the flagship of the Griptape pack for a reason: it bundles every driver OpenAI offers into a single CONFIG output, and one CONFIG wire is all an Agent needs to know what model to talk to.

How it works

The "Drivers" naming is the point. Instead of a single "structure config" that hard-codes one prompt model, this node builds a Griptape DrvsConfig with a separate driver for each capability. Under the hood it instantiates an OpenAI chat prompt driver, an image generation driver, an embedding driver, a text-to-speech driver, and an audio transcription driver, then bundles them into one CONFIG object. Wire that into the config input of any Griptape Agent (or Structure), and the agent picks up every one of them at once.

You don't need to configure all of them. Every field is optional; leave the image generation stuff alone and the agent simply won't generate images. That makes this a great "just make it work" node before you start fiddling with per-provider configs.

The inputs that matter

• model (default gpt-4o-mini) - the chat/prompt model. The dropdown lists 60+ OpenAI models, so you can drop in gpt-5 or gpt-4.1 without typing anything.
• api_key_env_var (default OPENAI_API_KEY) - the name of the env var or ComfyUI setting holding your key. Not the key itself.
• image_generation_model (dall-e-3 default, plus dall-e-2 and gpt-image-1) and size - for the image generation task nodes.
• embedding_model (text-embedding-ada-002, text-embedding-3-small, text-embedding-3-large) - used for vector store / RAG memory.
• text_to_speech_model + voice, and audio_transcription_model (whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe) - round out the audio side.

The shared knobs - temperature, seed, max_tokens, min_p, response_format (default vs json_object), max_attempts_on_fail - apply to the prompt driver. Two that trip people up: use_native_tools toggles between the LLM's own tool-calling and Griptape's tool engine, and max_tokens defaults to -1, which means "let the tokenizer decide" - leave it alone unless you're hitting context limits.

The only output is CONFIG, which feeds the config socket on an Agent or Structure node.

Installing it

You're installing the whole pack once:

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

or search "Griptape" in ComfyUI Manager and install from there. Either way it pulls in griptape[all] and python-dotenv - and a torch build, which can clash with your existing one; the README's troubleshooting section has the exact reinstall commands. Restart ComfyUI, then set your key in Settings → Griptape in the sidebar (or as an OPENAI_API_KEY env var - the node checks settings first, then the environment).

Common issues

The classic failure is a blank response and a console line like [WARNING]: OPENAI_API_KEY is not set. That means the node looked for the key and couldn't find it - check you entered it under Settings → Griptape and that the env var name you typed matches exactly. Also remember the pack installs griptape and torch into your ComfyUI environment; if ComfyUI breaks after installing this pack, that's almost always the torch version getting stomped - the README's StabilityMatrix and uninstall/reinstall steps are the fix.