Overview

Product video
Cognee is an AI memory engine. It ingests documents, extracts entities and relationships into a knowledge graph, embeds them for semantic search, and answers questions with graph- and passage-grounded retrieval.
The AWS CloudFormation template deploys into your existing VPC:
- the Cognee API on one Amazon EC2 instance (private by default; reached through your network, AWS Systems Manager port forwarding, or an optional internal HTTPS load balancer)
- Amazon Aurora PostgreSQL Serverless v2 for relational data, the knowledge graph, vectors and the session cache
- Amazon Bedrock for answers, and optionally for extraction and embeddings (Amazon Titan)
You choose the cost model at deploy time:
- GLiNER (Small Language Model): runs on the API instance's CPU, so the instance is larger, but ingestion makes no LLM calls for extraction.
- LLM: runs extraction on Bedrock instead.
Embeddings use Amazon Titan Text Embeddings v2 or a local FastEmbed model that ships inside the image.
Your data never leaves your account. Authentication and per-dataset access control are on by default, and telemetry is off.
Requirements:
- Amazon Bedrock model access for Anthropic Claude Sonnet 4.5, and for Amazon Titan Text Embeddings V2 if you use Titan embeddings.
- Subnets with a NAT gateway, or the VPC endpoints the template can create.
- AWS resources (EC2, Aurora, Bedrock usage) are billed separately by AWS.
Highlights
- Knowledge-graph memory for AI agents: graph and vector retrieval with LLM answers, all in your AWS account.
- One database: Aurora PostgreSQL stores graph, vectors, metadata and sessions, with point-in-time recovery and snapshots.
- Your choice of cost model: GLiNER extraction with no LLM calls, or Bedrock LLM extraction, plus Titan or built-in local embeddings.
Details
Features and programs
Financing for AWS Marketplace purchases
Pricing
Dimension | Description | Cost/12 months |
|---|---|---|
Cognee Deployment | One Cognee Python edition deployment in your AWS account. AWS infrastructure is billed separately by AWS. | $250,000.00 |
Vendor refund policy
Contract fees are billed through AWS Marketplace. To request a refund, contact support@topoteretes.com within 30 days of purchase, with your AWS account ID and the reason for the request. AWS infrastructure charges are billed by AWS and are not refundable by the seller.
How can we make this page better?
Legal
Vendor terms and conditions
Content disclaimer
Delivery details
Cognee on AWS - Python (CloudFormation)
Deploys a private, single-tenant Cognee AI memory server into your own AWS account and VPC. Your documents, knowledge graph and embeddings never leave your account.
What the template creates:
- One EC2 instance (default m7i.2xlarge; m7i.large is enough with LLM extraction) running the Cognee Python API from this AMI, with an encrypted EBS data volume, daily snapshots and snapshot-on-delete.
- Amazon Aurora PostgreSQL Serverless v2 (encrypted, private, TLS required) as the relational, graph, vector and session-cache store, with an optional reader instance for fast failover.
- Optional internal Application Load Balancer with TLS (ACM certificate) and Route 53 private DNS record.
- Optional VPC interface endpoints for subnets without internet access.
Model choices at launch:
- Graph extraction: GLiNER, a small local language model running on the API instance (no LLM calls for extraction, larger instance), or Amazon Bedrock Claude (pay per token).
- Embeddings: Amazon Titan Text Embeddings v2 or a local FastEmbed model included in the image.
Security: IMDSv2, encryption at rest, private database, no secrets in user data, least-privilege IAM, a required client CIDR (no public default), and a generated administrator password stored in Secrets Manager.
Requires Amazon Bedrock model access in the deployment region. AWS infrastructure is billed separately.
CloudFormation Template (CFT)
AWS CloudFormation templates are JSON or YAML-formatted text files that simplify provisioning and management on AWS. The templates describe the service or application architecture you want to deploy, and AWS CloudFormation uses those templates to provision and configure the required services (such as Amazon EC2 instances or Amazon RDS DB instances). The deployed application and associated resources are called a "stack."
Version release notes
Fixes a launch failure in 1.0.0: the API server's root disk is now 40 GB, the size of the AMI's root snapshot (1.0.0's template asked for 30 GB, which EC2 rejects). Same disk and software as 1.0.0: the Cognee Python server (cognee 1.6.2 with the Postgres graph adapter) on Amazon Linux 2023 with Aurora PostgreSQL Serverless v2 and Amazon Bedrock.
Additional details
Usage instructions
Launch with the provided CloudFormation template; launching the AMI by itself does not start a working deployment.
- Choose your VPC (VpcId), one private subnet and its Availability Zone for the API (SubnetId, AvailabilityZone), and at least two private subnets in different Availability Zones for Aurora (DBSubnetIds). Subnets need a NAT gateway, or set CreateVpcEndpoints=true and provide S3RouteTableIds.
- Set AllowedCidr to the network that will call the API (required, no default) and AdminEmail for the first administrator account. AssignPublicIp=true requires EnableTlsLoadBalancer=true.
- Choose graph extraction: "gliner" (default; a small local model on the API instance, no LLM calls for extraction; keep the default m7i.2xlarge or larger) or "llm" (Amazon Bedrock Claude, pay per token; m7i.large is enough). The stack outputs the estimated hourly instance cost.
- Choose embeddings: "titan" (Amazon Bedrock Titan v2, 256/512/1024 dimensions) or "fastembed" (local model included in the image, 384 dimensions). This cannot be changed after the first ingestion without a full re-index.
- Enable Bedrock model access for Anthropic Claude Sonnet 4.5 and, if used, Amazon Titan Text Embeddings v2 in the deployment region.
- After CREATE_COMPLETE, allow several minutes for the first start (database migrations), then read the API URL and the port-forward command from the stack outputs (ApiUrl, PortForwardCommand). The administrator password is in Secrets Manager (AdminSecretArn output); change it at first login.
External dependencies (billed by AWS separately): Amazon Bedrock, Aurora PostgreSQL Serverless v2, Application Load Balancer and ACM (TLS only), Secrets Manager, CloudWatch Logs, EBS snapshots.
Documentation: https://docs.cognee.ai
Resources
Vendor resources
Support
Vendor support
Cognee provides support for this product by email and through our community channels.
Email: support@topoteretes.com Documentation: https://docs.cognee.ai GitHub issues: https://github.com/topoteretes/cognee/issues Community (Discord): https://discord.gg/NQPKmU5CCg
Support covers deployment of the CloudFormation template, configuration of the Cognee server, and product defects. We respond to email requests within 2 business days.
When you contact us, please include your AWS Region, the product version, the stack parameters you changed, and any relevant CloudWatch logs. Do not send secrets or customer data. We have no access to your AWS account or instances unless you explicitly grant it.
AWS infrastructure and Amazon Bedrock issues, including quotas and model access, are handled through AWS Support.
AWS infrastructure support
AWS Support is a one-on-one, fast-response support channel that is staffed 24x7x365 with experienced and technical support engineers. The service helps customers of all sizes and technical abilities to successfully utilize the products and features provided by Amazon Web Services.