Costs and Dependencies
How DevStride tracks what it costs to run, which services are billed by use, the main software the product is built on, and how dependencies are kept up to date and safe.
On this page
Costs and Dependencies#
How running costs are tracked#
DevStride staff have a Cloud costs page in the product, under Settings → DevStride Admin, which shows the cost of running DevStride. A scheduled job refreshes it every day from each provider's billing API, with nothing entered by hand.
| Provider | What is read |
|---|---|
| AWS | Both AWS accounts (production and non-production), by day, by service and by environment. The CI machines are shown as their own line. |
| Neon | Database compute, by day |
| GitHub | Usage at list price, with any credits shown separately |
| Vercel | Website hosting charges |
The page shows the month to date with a projection for the month end, the trend for each provider, each account broken down by service, and the month broken down by service or by environment. Every AWS resource is tagged with the environment it belongs to, so costs can be split by environment. Engineers can see the same figures from the command line, including the cost of each development machine and environment.
AI usage. The cost of AI features is tracked inside the product for each organization, with a monthly spending limit for each organization and alarms on usage.
What the page does not include. Services billed outside these four providers, such as Pusher, Stripe's payment fees, the AI providers' own invoices, Slack and the AI coding tools, are tracked separately from the product.
Services billed by use#
Most of DevStride's running cost rises and falls with use rather than being fixed:
| Service | What drives the cost |
|---|---|
| AWS Lambda, API Gateway, SQS, SNS, EventBridge | Number of requests and events, and how long functions run |
| Amazon S3 and CloudFront | Stored files and data served |
| Amazon SES | Emails sent and received |
| Amazon CloudWatch | Logs stored, metrics and alarms |
| Neon | Database compute time and storage |
| Pusher | Connections and messages |
| OpenAI and Anthropic | AI requests and embeddings |
| Stripe | A share of each payment |
| GitHub and RunsOn | CI minutes and machines |
Any change that adds a new billed resource, such as a new service, a storage bucket with long retention, or a new set of alarms, is agreed with the owner before it ships.
What DevStride is built on#
| Layer | Main components |
|---|---|
| Language and runtime | TypeScript 5.7, Node.js 22 |
| Package manager | pnpm 11, as one workspace |
| Infrastructure | SST 2.47, AWS CDK 2.171 |
| Backend | Hono 4, Zod 3, Drizzle ORM, node-postgres, AWS SDK for JavaScript v3, AWS Lambda Powertools 2 |
| Web app | Vue 3.5, Quasar 2.23, Pinia 2, Vue Router 4, Tailwind CSS 3, Vite 6, TipTap 3, AG Grid, Chart.js, d3 |
| Real-time, billing and AI clients | Pusher, Stripe, the Vercel AI SDK with its OpenAI and Anthropic providers, Slack Web API |
| MCP server | The official Model Context Protocol SDK |
| Testing | Vitest 4, Vue Test Utils |
Keeping dependencies safe#
- Every licence checked. CI checks every dependency's licence against an approved list. Any exception is recorded.
- A 24-hour waiting period. No package version can be installed until it has been public for at least 24 hours, which protects against freshly published malicious releases. Packages pulled from unusual sources, such as git URLs or tarball links, are refused.
- Install scripts allowed one by one. Only packages on an explicit allow-list may run scripts when they are installed.
- Known vulnerabilities pinned past. Known vulnerable versions deep in the dependency tree are overridden to safe ones, and the workspace is held to a single copy of key libraries.
- Weekly updates. Dependabot proposes updates weekly, grouped and held back a few days after release, for review through the normal process.
- Every change scanned. Every change is scanned for secrets and for common security mistakes before it merges.