Connect Cognee to AWS Bedrock — Keep Your AI Data Inside AWS

You've built a knowledge graph with Cognee locally. You've promoted it to Cognee Cloud. Now what if your company runs everything through AWS — committed spend, IAM-based access control, CloudTrail audit logs, and a security team that doesn't want a single byte leaving the AWS boundary?

In this video, Brandon will show you how to connect Cognee to Amazon Bedrock so every LLM and embedding call runs through your existing AWS account — billed to your AWS invoice, controlled by IAM, and audited by CloudTrail. We'll use Claude Sonnet 4.6 for inference and Amazon Titan for embeddings, all wired up through LiteLLM Proxy.

The best part? Your Cognee code doesn't change at all. It's still cognee.remember() and cognee.recall() — LiteLLM acts as the adapter between Cognee's OpenAI-compatible interface and Bedrock's API on the back end.

🔗 What we cover:
00:00 Why run Cognee on AWS Bedrock?
01:05 What Bedrock gives you: billing, IAM, CloudTrail, top-tier models
02:17 How Cognee talks to Bedrock through LiteLLM Proxy
03:14 Installing cognee[aws] and litellm[proxy] with uv
03:48 Creating the LiteLLM config file
05:25 Finding model IDs and inference profiles in the AWS console
06:49 Creating an IAM user and access keys for Bedrock
07:41 Wiring credentials and model IDs into the config
08:19 Setting up .env for Cognee → LiteLLM → Bedrock
09:54 Running LiteLLM Proxy locally on port 4000
11:06 Ingesting docs and querying the knowledge graph through Bedrock
11:51 ⚠️ Production credentials: use IAM roles, not keys in YAML
12:32 Recap of what we built

🛠️ Tools used:
Cognee (cognee[aws])
LiteLLM Proxy (litellm[proxy])
Amazon Bedrock — Claude Sonnet 4.6 + Amazon Titan Text Embeddings v2
Python 3.12 + uv
AWS IAM for access control

⚠️ Security note: The demo puts AWS keys directly in the LiteLLM YAML for local development. Never commit that file to Git. In production, assign an IAM role to your EC2 instance, ECS task, or Lambda function — LiteLLM will pick up credentials automatically via the instance metadata service. The minimum required permission is bedrock:InvokeModel.

▶️ Watch the rest of the series first:
Part 1 — Building a knowledge graph with the Cognee SDK: [link]
Part 2 — Sharing your knowledge graph with Cognee Cloud: [link]

This video is sponsored by Cognee.
The result: same Cognee code, same high-quality knowledge graph, but now every call flows through your AWS account — perfect for enterprises with AWS commitments or strict compliance requirements.
👍 Like and subscribe for more videos on Cognee, knowledge graphs, and LLM-powered developer tools.
#AWS #Bedrock #Cognee #KnowledgeGraph #Python #Claude #LiteLLM #LLM #AI #DeveloperTools

Видео Connect Cognee to AWS Bedrock — Keep Your AI Data Inside AWS канала cognee