Operator guide for Valet Dedicated 0.10.1

Cost and sizing

This page estimates what an installation costs in AWS and what makes the bill grow. Every figure is an estimate. It comes from AWS on-demand list prices for us-east-1, 730 hours a month, and the module's default sizes. Valet has not yet measured a running installation against it. Check the figures in the AWS Pricing Calculator for your region before you budget.

What this page leaves out

The estimates cover the resources the module creates. They do not include:

  • Your VPC's NAT gateways and interface endpoints. The installation runs in your VPC and uses what it has. Outbound traffic through your NAT is billed to you; see what grows the bill.
  • Model usage. Anthropic bills your API key, and OpenAI does when you set an OpenAI key.
  • Your WorkOS and identity provider plans.
  • Valet's fee, which your agreement sets.

An empty installation

An empty installation runs no agents or apps. These are the module's defaults: one task per service, and a single-zone database.

Platform tasks

What it is
api, worker, and inference at 0.5 vCPU and 1 GB; dashboard and build at 0.25 vCPU and 0.5 GB; one task each
Estimate per month
$72

Standby runtimes

What it is
2 runtimes at 0.25 vCPU and 1 GB, the Dedicated default for runtime.standby_count
Estimate per month
$21

Platform database

What it is
RDS PostgreSQL 17, db.t4g.small, 20 GB, one zone
Estimate per month
$26

Internal load balancer

What it is
The hourly charge and about one load balancer capacity unit
Estimate per month
$22

Storage and backups

What it is
The EFS file system for source code, S3 buckets, ECR images, and AWS Backup copies, all near empty
Estimate per month
$3

Secrets Manager

What it is
10 secrets
Estimate per month
$4

DNS

What it is
The private hosted zone
Estimate per month
$1

Logs

What it is
CloudWatch Logs ingestion at idle
Estimate per month
$2

Total

What it is
Estimate per month
about $150

The installation has no public IPv4 address, no public load balancer, and no web application firewall, so the estimate has no charge for them.

Container Insights is off. Turning it on adds about $1.40 a day per cluster.

What grows the bill

Most costs grow with one driver each.

Running agents and apps

Unit cost
$0.0146 per runtime-hour (0.25 vCPU, 1 GB). About $10.60 a month for a runtime that never idles
Example
50 agents active 8 hours each weekday: about $130 a month. 20 apps that never idle: about $210

Builds

Unit cost
About $0.10 per builder-hour (2 vCPU, 4 GB, 50 GB disk). A 5-minute build costs about $0.01
Example
1,000 builds: about $10

Outbound traffic

Unit cost
$0.045 per GB through your NAT gateway, plus internet transfer fees past the free tier
Example
Package installs, browsing, and connector calls at 500 GB: about $23 plus transfer

Platform database

Unit cost
Storage grows at $0.115 per GB-month. A larger class costs more
Example
db.t4g.medium is about $47 a month

Request load

Unit cost
About $18 per added api task
Example
Two more api tasks: about $36

Logs

Unit cost
$0.50 per GB ingested, then $0.03 per GB-month stored
Example
20 GB of agent and app logs a month: about $10

Source and artifacts

Unit cost
EFS at $0.30 per GB-month; S3 at $0.023 per GB-month
Example
50 GB of repositories: about $15

Orgs

Unit cost
Certificates and DNS records for orgs are free
Example
About 24 orgs fit; see Limits

Runtimes dominate growth. An idle runtime costs nothing. When an agent claims a standby runtime, the installation starts a replacement, so the standby pool stays the same size.

Choose the sizes

The module takes its sizes from these inputs. The reference lists each one.

sizes

Default
One task per service, at the sizes in the table above
What it changes
CPU, memory, and task count per service. More api tasks serve more requests.

database

Default
db.t4g.small, 20 GB
What it changes
The platform database's class and storage

app_database

Default
Off
What it changes
A second RDS instance for app databases; see below

backup

Default
On, 35 days
What it changes
How long AWS Backup keeps each recovery point

The runtime.standby_count setting sets the standby pool, 2 by default. Each standby runtime costs about $10.60 a month. A larger pool starts agents faster and costs more; see Limits.

The module runs the database in one zone. A second task per service doubles that service's cost and the database connections it opens. A db.t4g.small cannot hold the connections of two tasks per service, so raise database.instance_class before you raise a count.

The app database

Turning on app_database adds a second RDS instance. It costs about $26 a month at the defaults, db.t4g.small with 20 GB, and the cost grows with its class and storage. Every app database shares the instance. See App database.

Log retention

Each log group the module creates keeps its events for log_retention_days, 365 by default. The audit log group and the database log group keep them for audit_retention_days, 400 by default. CloudWatch Logs charges storage by the GB-month for as long as an event stays. A longer window costs more once the logs reach it. At steady state, 20 GB of logs a month stored for 365 days is about 240 GB, about $7 a month. Audit events and database connection records are small. Shorten log_retention_days to cut storage cost. Lowering a value deletes older events when you apply. See Observe.

AWS Backup keeps each recovery point for backup.retention_days. Backups of the database and the file system grow with their data.

Use the numbers

Take the empty installation as a floor. Add a runtime estimate for your expected agents and apps, and keep a margin. These figures are estimates until measured on a running installation.