Griptape Embedding Driver: Amazon Bedrock Titan

Embedding drivers are the unsung plumbing of the Griptape pack. They don't make anything visible - they turn text into the vectors that power RAG, memory search, and vector-store lookups. Griptape Embedding Driver: Amazon Bedrock Titan is the version that produces those vectors with Amazon's Titan embedding models, billed to your AWS account.

Reach for it when your whole stack is already AWS: the credentials you set up for the Bedrock image drivers work here too, and your embedding traffic stays inside the same cloud contract. If you don't have AWS, it's the wrong driver - there are far easier options in the pack (Ollama, LM Studio, OpenAI).

How it works

Like every driver in the pack, this node is configuration, not computation: you set the parameters and it produces an EMBEDDING_DRIVER object. That object plugs into the things that actually embed text - typically a vector-store driver or a RAG-related node in the pack that takes an embedding driver. The node itself calls nothing until something downstream uses it.

The inputs:

• embedding_model - three Titan options: amazon.titan-text-premier-v1:0 (default), amazon.titan-text-express-v1, and amazon.titan-text-lite-v1. Premiere is the strongest but pricier; lite is the cheap, fast option.
• min_retry_delay / max_retry_delay - retry backoff bounds for rate-limited or flaky calls. The tooltip on max says "Minimum retry delay" but it's the upper bound; sensible defaults of 2 and 10 seconds are fine.
• aws_access_key_id_env_var / aws_secret_access_key_env_var / aws_default_region_env_var - the pack's standard AWS trio, taking env-var names, not keys.

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 torch caveat applies pack-wide (see the README: reinstall torch with the CUDA index if Griptape's install breaks ComfyUI's build).

Gotchas

Most problems are Bedrock access problems in disguise: if Titan embedding isn't granted in your region, you'll get an access error from the API even though the node is configured perfectly - grant the model under Bedrock → Model access. Don't overthink the retry delays unless you're hitting throttling; AWS rate-limits embedding calls, and if you're churning through a big document set you'll meet them. And keep the model choice honest: lite vs. premiere changes vector quality and cost, so match the model to how much precision your search actually needs.