Griptape Agent Config: Expand

The name is the whole trick: this node takes a Griptape CONFIG - the bundled blob of drivers that config nodes like "Griptape Agent Config: OpenAI Drivers" hand you - and unpacks it into the six individual drivers inside. One wire in, six wires out. It's the node you reach for when a config almost does what you want, but you need to reroute one piece of it.

Why you'd use it

Config nodes output one fat CONFIG object containing a prompt driver, an image generation driver, an embedding driver, a vector store driver, a text-to-speech driver, and an audio transcription driver. Usually that's exactly what an Agent wants. But sometimes you don't want the bundle:

You want to take the prompt driver from one config and pair it with a different embedding or vector store driver.
You want to inspect what a config actually contains before committing to it.
You're building a driver-level workflow and want to wire a specific driver into a task node directly, skipping the Agent's config socket entirely.

That last one is the real use case. Griptape task and driver nodes accept individual drivers, so "Expand" is how you bridge between the config world and the driver world without rebuilding everything by hand.

How it works

The mechanism is nearly embarrassing in its simplicity. The node takes the config input, reads the six driver attributes off it - prompt_driver, image_generation_driver, embedding_driver, vector_store_driver, text_to_speech_driver, audio_transcription_driver - and returns them as six separate outputs. If you pass it an empty or invalid config it raises a "No config provided" error rather than silently returning nothing, which is good news for debugging. The source marks it as an output node, so it behaves like a terminal in the graph.

Inputs and outputs

The single required input is config (CONFIG). Connect any of the pack's config nodes here - OpenAI Drivers, Google Drivers, Ollama Drivers, the deprecated ones, all of them output the same type.

The six outputs, in order:

PROMPT_DRIVER - the LLM chat driver
IMAGE_GENERATION_DRIVER - drives the image generation tasks
EMBEDDING_DRIVER - for vector store / RAG work
VECTOR_STORE_DRIVER - the memory store, usually local
TEXT_TO_SPEECH_DRIVER - for the text-to-speech task nodes
AUDIO_TRANSCRIPTION_DRIVER - for the audio transcription task nodes

There are no other inputs or knobs to fiddle with. You feed it a config, it does the rest.

Installing it

Part of the one-pack install. ComfyUI Manager → search "Griptape" → install, or:

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

Then restart ComfyUI. The pack brings in griptape[all] and python-dotenv automatically; be aware griptape also installs a torch build into your environment, which is the usual source of post-install breakage (the README has the uninstall/reinstall fix for Nvidia).

Common issues

There's genuinely not much to go wrong here - the node is a straight unpack. The failure you're most likely to see is upstream: if the config node feeding it couldn't build its drivers (usually a missing API key or a mistyped env var name), you'll get the "No config provided" error or a None result. Check the key setup before suspecting this node. Also note that not every provider config carries every driver - for instance, the Groq config only bundles prompt + audio transcription, so the image generation output will be empty on that path.