Griptape Agent Config: OpenAI Compatible [DEPRECATED]

The deprecated OpenAI-compatible config is the older, less capable version of the node you should actually be using. It does the same headline thing - point Griptape at any OpenAI-compatible API by changing a base_url - but it's a generation behind the "Drivers" configs, and you'll mostly meet it inside old workflows. Let's be clear about what you lose and why you should switch.

What it is

A single config node with a base_url (default https://api.openai.com/v1) that lets you aim a Griptape agent at any OpenAI-compatible server. Optional inputs cover prompt_model (default gpt-4o), image_generation_model (default dall-e-3), text_to_speech_model (default tts-1), plus api_key_env_var, temperature, seed, max_attempts_on_fail, env, use_native_tools, and max_tokens. Output is a single CONFIG.

The mechanism matches the modern node: it builds OpenAI-style drivers and points them at your chosen base_url. That part never went away.

Why it's deprecated

The modern "Griptape Agent Config: OpenAI Compatible Drivers" is a strict upgrade on the same idea. Where the old node's fields are mostly free-text strings, the new one gives you proper dropdowns (models, sizes, voices, transcription models), adds an embedding driver and audio transcription driver that this old version simply doesn't have, and generally matches the newer per-provider config architecture. The old node will keep loading legacy workflows, but it's not where you want to be.

Installing it

It ships in the one pack:

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

or ComfyUI Manager → search "Griptape". Restart, set the appropriate API key in Settings → Griptape (or as an env var), and old workflows will load.

Should you use it?

If you have an existing workflow that uses it and it works, leave it until you next touch that workflow - then migrate. The swap is genuinely low-effort: the new node occupies the same CONFIG socket and defaults to the same https://api.openai.com/v1 base URL, so you re-enter your model names and you're done. For new work, don't add this node deliberately; the Drivers version is the one you want, especially if you also need embeddings or audio.

Common issues

• No embedding / audio support - expected with the old design; if your workflow needs RAG or transcription, that's a reason to migrate now rather than later.
• Model not found at the target server - the free-text prompt_model must match what the server actually serves (LM Studio's loaded model, vLLM's served model, etc.).
• [WARNING]: ... is not set - the standard key-missing console message; check Settings → Griptape and env var spelling.