Cost estimate¶

The Cost tab estimates monthly cost from the public AWS Price List (on-demand list prices). It is deterministic code, never a model's arithmetic, and every number can be traced to a rate, a quantity and an assumption.

The Cost tab

What the tab shows¶

Totals per month and per year, how many components were priced, and how many were not.
By category (compute, data and storage, AI and agents, …) and by component, with every line item expanded as quantity × rate = cost and the price-list SKU.
Assumptions — every input is tagged from spec (you stated it) or assumed (a default). Any defaulted input makes the estimate indicative.
Sensitivity at 1×, 3× and 10× traffic, and what-ifs (Lambda on arm64, RDS Multi-AZ cost, Bedrock batch inference, S3 Infrequent Access) where they apply.
Provenance: the price-list publication date for each service and the retrieval date.
A checkbox that overlays $/mo on the diagram.

Giving it real numbers¶

Put a usage object on each node. Anything you leave out falls back to a documented default and is flagged as assumed.

{ "id": "fm", "icon": "bedrock", "label": "Amazon Bedrock",
  "usage": { "model": "Claude Sonnet 5.5", "inputTokensPerMonth": 150000000, "outputTokensPerMonth": 15000000 } }

Describe the scenario once in meta.cost.note so readers know the basis:

"meta": { "title": "…", "cost": { "region": "us-east-1", "note": "Illustrative: 200,000 questions/month, ~750 input and 75 output tokens each" } }

Usage keys by service¶

Service	Keys (default)
AWS Lambda	`requestsPerMonth` (1e6), `avgDurationMs` (200), `memoryMb` (512), `arch` (`x86` or `arm64`)
Amazon API Gateway	`type` (`rest` or `http`), `requestsPerMonth` (1e6)
Amazon DynamoDB	`readRequestsPerMonth` (5e6), `writeRequestsPerMonth` (1e6), `storageGb` (10)
Amazon S3	`storageClass` (`standard`), `storageGb` (100), `putRequestsPerMonth` (1e5), `getRequestsPerMonth` (1e6)
Amazon SQS	`requestsPerMonth` (3e6), `fifo` (false)
Amazon SNS	`publishesPerMonth` (1e6)
AWS Step Functions	`stateTransitionsPerMonth` (1e6)
Amazon EventBridge	`eventsPerMonth` (1e6)
Amazon CloudWatch	`logIngestGbPerMonth` (10), `logStorageGb` (10), `customMetrics` (20), `alarms` (10)
AWS KMS	`keys` (1), `requestsPerMonth` (1e5)
AWS WAF	`webAcls` (1), `rules` (5), `requestsPerMonth` (1e6)
Amazon Cognito	`tier` (`essentials`), `mau` (1000)
AWS Secrets Manager	`secrets` (5), `apiCallsPerMonth` (1e5)
Amazon Textract	`feature` (`text`, `tables`, `forms`…), `pagesPerMonth` (10000)
Amazon Comprehend / Comprehend Medical	`feature` (`DetectEntities`), `unitsPerMonth` (1e6)
Amazon OpenSearch Service	`mode` (`provisioned` or `serverless`), `instanceType` (`r6g.large.search`), `instanceCount` (2), `storageGbPerNode` (100); serverless: `indexingOcu`, `searchOcu`, `storageGb`
AWS Fargate	`tasks` (2), `vcpu` (1), `memoryGb` (2), `hoursPerMonth` (730), `arch`
Amazon EC2	`instanceType` (`m5.large`), `count` (2), `hoursPerMonth` (730), `ebsGbPerInstance` (50)
Amazon RDS	`instanceClass` (`db.r6g.large`), `engine` (`PostgreSQL`), `multiAz` (true), `storageGb` (100)
Amazon Aurora	`instanceClass`, `engine` (`Aurora PostgreSQL`), `instances` (2), `storageGb` (100)
Elastic Load Balancing	`type` (`application`), `avgLcu` (5)
NAT gateway	`count` (2), `gbProcessedPerMonth` (100)
Bedrock Guardrails	`textUnitsPerMonth` (2e5), `policies` (`["content","pii"]`, also `grounding`)
Amazon Bedrock (models)	`model` is required (for example `Claude Sonnet 5.5`, `Nova 2.0 Lite`, `Titan Embedding V2`), `mode` (`standard`, `batch`, `global`), `inputTokensPerMonth` (5e6), `outputTokensPerMonth` (1e6), `cacheReadTokensPerMonth`
Bedrock AgentCore	`component` (`runtime`, `gateway`, `memory`, `codeinterpreter`, `browser`, `evaluations`; detected from the label when omitted) plus per-component volumes such as `sessionsPerMonth`, `avgSessionMinutes`, `vcpu`, `memoryGb`, `cpuActiveFraction`, `invocationsPerMonth`

Two keys work on any node:

usage.monthlyUsd — supply the monthly cost yourself (status override). Use it for services without a cost model, and say where the number came from.
usage.egressGbPerMonth — adds internet data-transfer-out.

Honest statuses¶

Status	Meaning
`estimated`	Priced from the price list with the stated and assumed inputs.
`override`	You supplied `usage.monthlyUsd`.
`no-charge`	The service has no direct charge (IAM Identity Center, Organizations, CloudFormation…).
`not-billable`	A general icon (users, internet), not an AWS resource.
`not-itemized`	AWS lists no separate price for it (some AgentCore components bill through CloudWatch or the runtime).
`needs-input`	A required input is missing (for example the Bedrock `model`).
`not-estimated`	No cost model for this service yet. Supply `usage.monthlyUsd` to include it.
`error`	The price book could not resolve an unambiguous rate; the reason is shown.

The total only includes priced components, and the tab says how many were left out.

What is and is not included¶

Included: public on-demand rates; 730 hours per month; binary storage units. Excluded: free tier, taxes, support plans, Savings Plans, Reserved Instances, Spot, enterprise discounts, credits, and data transfer unless you set egressGbPerMonth. Treat the result as a planning number and confirm it in the AWS Pricing Calculator.

From the command line¶

archify-aws cost spec.json                       # table in the terminal
archify-aws cost spec.json --scale 1,3,10        # sensitivity points
archify-aws cost spec.json --region us-east-1    # needs data/prices/<region>.json
archify-aws cost spec.json --usage usage.json    # {"<nodeId>": {…usage keys…}} overrides the spec's usage
archify-aws cost spec.json --json                # full machine-readable result

Turn the tab off with render --no-cost / finalize --no-cost.

The price snapshot is committed (data/prices/us-east-1.json, 26 services). Developers refresh it with npm run prices:fetch; see Data refresh.