Before you start
What you need ready before you launch the eode stack.
From River Logic
Ask your River Logic contact to invite your 12-digit AWS account ID. Once you're invited, your contact sends you:
- the Launch Stack link for the current release;
- your sign-in details (client ID, sign-in domain, API key, webhook secret), after your company login is connected to the sign-in service. See Sign-in and access.
In your AWS account
| You need | Why |
|---|---|
| An administrator role that can create IAM roles, EKS, RDS, ElastiCache and EFS, and a VPC unless you use your own | The stack creates them |
| On-demand vCPU quota of at least 32 in the region (the standard instance families) | New accounts start at 5; ask AWS Support to raise it before you launch |
Two hostnames you control, for example eode.example.com and eode-mcp.example.com | The web app and the MCP endpoint |
| An Anthropic API key from your own Anthropic account, with a spend limit set there | Every conversation with RIA uses it |
Decide before you launch
-
New VPC or existing? The stack can create a VPC, or install into yours. It always creates a new EKS cluster; to use a cluster you already run, see Install with Helm. An existing VPC needs:
- DNS support and DNS hostnames turned on;
- two or three private subnets in different availability zones, with outbound internet access (for example through NAT), for the cluster, database, cache and file storage;
- public subnets with an internet gateway, tagged
kubernetes.io/role/elb=1, for the internet-facing load balancer.
In an existing VPC the stack creates no subnets, routes, gateways or endpoints. It does create security groups for the database, cache, file storage and installer, the database and cache subnet groups, and the file system's mount targets in your subnets. It never creates DNS records; you create the two CNAMEs yourself. See Release status.
-
Certificates: automatic (Let's Encrypt, which needs your hostnames to reach the stack from the internet on ports 80 and 443) or your own certificate in AWS Certificate Manager, covering both hostnames.
-
Size: each active user's engine reserves half a CPU and 2 GiB of memory, and can grow to its memory size (12 GiB by default). One default engine node (
r6i.2xlarge, 8 CPUs, 64 GiB) holds about 14 engines while models are small, and about 5 if every user works on a large model at once. Raise EngineNodeCount for more. See Settings. -
Cluster access: set AdminPrincipalArn to the IAM role your platform team uses. Without it, nobody in your account can use
kubectlon the cluster.