Design Mode draws your infrastructure. Live Mode reads what you already shipped. Point it at an SST or Vercel repo. It rebuilds the diagram, scores its confidence, and shows the data flow. It runs on your laptop. It uses no credentials and makes no network calls.
$ node scripts/sst-dream.mjs scan .
reading sst.config.ts + packages/infra/*.ts
redacting secrets before parse
recovered 14 resources · 9 edges
✓ wrote ARCHITECTURE.md + sstdream-scan.json
not recognized: 1 (listed, never dropped)
$ Most observability tools hand you a dashboard that looks nothing like how you picture your system. SSTDREAM already has your shape. You drew it in Design Mode, imported it from code, or loaded it from the gallery. Live Mode runs the same engines over your project on disk: validation, data-flow simulation, cost, expansion, and the security audit. Same picture, real source. No second mental model to learn.
sst-dream scanPoint it at a repo. Get a straight answer.
This part is real and ships today as a local command. It strips secrets before the code reaches the parser. It reverse-parses the project. It runs the builder's engines over the recovered graph. Then it tells you exactly what it could and could not model.
No clone. No install. No build. Copy the scripts/ folder into your SST or Vercel project and run one file.
# copy the repo's scripts/ folder into your project, then:
cd your-sst-project
node scripts/sst-dream.mjs scan . # or: ./scripts/sst-dream.sh scan .
# writes ARCHITECTURE.md + sstdream-scan.json here (or --out <dir>)
# no credentials, no network, nothing uploadedWant the visual version? Open the builder, click From code, and paste the sstdream-scan.json. It loads as an editable canvas. Details are in docs/live-mode.md.
SSTDREAM is a static site on GitHub Pages. The site shows off Live Mode. It does not run it. The scan runs from the repo you cloned, on your own machine, never from the hosted URL. Your code today, and your read-only cloud credentials when the observed view lands, stay on your laptop. There is no SSTDREAM backend to send them to. There will never be a connect-your-AWS-to-sstdream.com button. Local-first is the point. You can read the source to confirm it.
Plainly: this part is not built yet. The plan closes the loop. Connect read-only AWS or Vercel from your machine. Live Mode pulls your real signals onto the diagram you already drew. Which node is hot. Which edge is slow. Where the spend sits. Where the deployed reality has drifted from your code. It is the X-Ray view, except it is free, local, and pinned to your design instead of a generic service map.
X-Ray, Datadog, and Cloudcraft are strong tools. They also route your telemetry and your bill through someone else's platform. Live Mode makes a different trade on purpose. It is free, open source, and runs on your machine. You do not pay per seat or per host to look at your own architecture. For understanding the system you designed, local and free wins.
Bring your own model. The agent is local, read-only, and grounded in your scan. Today it does one concrete thing. It flags deprecated SST deterministically with agent check. It reads. It does not deploy and it does not write infra.
Live Mode only adds. It is a second pillar on its own route. The builder, the gallery, From code import, and verified export all work exactly as before. The builder is still zero-AI, zero-network, deterministic, and credential-free. Live Mode reads your project. It never feeds the generator.