Point Skeram at a repo. It maps the actual code into an interactive AWS architecture — then drag from 100 to 100,000 users and watch costs, bottlenecks, and failure points shift. Deterministic numbers with the formulas shown, not chat-bot vibes.
Real open-source apps, run through the real pipeline. Every page is live — drag the slider, click the nodes, audit the formulas.
Open-source DocSend alternative — Next.js, Prisma, Postgres, S3 uploads. The classic document-sharing SaaS shape.
Link shortener with a redirect hot path, Upstash Redis, QStash jobs, and Tinybird analytics. Reads dominate everything.
The archetypal AI chat app — Supabase everywhere, streaming LLM calls held open in the request path.
Open-source survey platform — 84 routes, webhooks, crons, integrations, and email jobs hanging off one Next.js app.
Self-hosted web analytics — every tracked pageview is a public POST to /api/send. Postgres + optional ClickHouse/Kafka at scale.
A meal-planning SaaS analyzed from its design doc alone — no repo yet. Skeram models the architecture before the first commit.
Golden-path stacks first: the pipeline speaks the JS/TS ecosystem — Next.js first-class; Express, Fastify, and Hono best-effort. Other stacks are politely declined rather than guessed at.
The LLM reads the code once. Everything you see — every dollar, every red node — is computed deterministically from a typed model. Same repo, same numbers.
Claude maps routes, data access, queues, jobs, and external calls into a schema-validated AppModel. Every claim carries evidence — the file and lines it came from. No evidence, no claim.
A curated AWS pricing table and a failure-pattern rule catalog run over the model in pure TypeScript. Costs, saturation, and findings are math you can audit — click any number to see its formula.
The diagram re-renders per scale: nodes resize, costs recompute, bottlenecks glow ember. Next: export the CDK that deploys the right architecture into your own AWS account.
The short version. The full methodology — the schema, the eval bench, the human learning loop — is at /how-it-works.
Ask a chatbot about your architecture and you get prose. Skeram gives you:
LLMs can’t do reliable arithmetic across pricing SKUs. Skeram’s cost engine is hand-curated pricing data plus unit-tested sizing formulas — hallucination-free dollars.
Not a transcript — a shareable page. Every analysis is a URL you can send, embed, and (soon) keep updated on every push.
The endgame is generated CDK — validated, right-sized to your chosen scale, deployed into your own account. Your code, your infra, eject anytime.
The oracle is open. Sign in with GitHub, get five analyses for $10, and point Skeram at any public repo — you get a living page like the ones above, private to you until you choose to share it.
Prepaid credits, no subscription · private-repo support via the GitHub App is next