System requirements

Construct runs on a modest laptop. What you can do depends on how much memory, how many processor cores and how much free disk the machine has, and Construct tells you exactly where your machine stands instead of failing halfway.

Check your machine#

construct doctor

construct doctor reads your machine and prints the tier it meets, why, what is switched on and off and why, and the exact command that fixes each missing optional item. It changes nothing, uses no network beyond a 2 second look for Ollama on your own machine, and starts no model. It exits with 0 whatever it finds, except when your Node is too old or the machine cannot be read (exit 1). Add --format json for the same report as data.

The three tiers#

Tier	For	Cores	Memory	Free disk
Lite	the command line only: summarize, validate, chains and plans, no Cockpit, no model	2	4 GB	about 1 GB
Cockpit use	the prebuilt Cockpit, or a hosted one used from a browser, on top of the command line	2	8 GB	about 2 GB
Contributor	building and testing this repository: the full test suite, the Cockpit build, Playwright	4	8 GB workable, 16 GB comfortable	about 6 GB

A machine that reports a little less memory than its label (a "16 GB" laptop often shows 15 GB, because the system keeps some) still counts: a tier's memory need is met at nine tenths of the number. Construct needs Node 20 or newer at every tier.

• Lite is enough to install Construct, create and check a project, summarize it, run chains and plans, and let a language model drive the command line. Nothing here starts the Cockpit or loads a model.
• Cockpit use adds the browser app, either the prebuilt one or a hosted one you open from a browser. You never build the Cockpit yourself at this tier.
• Contributor is for working on Construct itself: the full test suite, the Cockpit build and the browser tests. With less than the comfortable amount of memory, run one heavy job at a time (packages/tools/dev/heavy.sh npm test waits for the machine to be free enough before it starts).

What a small machine switches off#

On a Lite machine, or anything below it, every choice is made by the rules that ship with Construct. No plugin is loaded, no model is downloaded and nothing is slowed down waiting for one: the command line says "rules only" and the reason, and carries on. The features that need more than the command line are decided from the machine, not from a model:

Feature	Needs
Model-backed proposals (a project's decision plugin)	Cockpit use or above
The Cockpit	Cockpit use or above
Playwright proofs (a screen proved in a browser)	Cockpit use or above, and a Playwright browser
Studio voice	Cockpit use or above, ffmpeg and a local voice

Each one has a rules-only fallback that always works.

Optional tools#

construct doctor checks for these and prints the fix line when one is missing: ffmpeg and ffprobe (Studio video), a Playwright browser (npx playwright install chromium), Ollama (a local model server, looked for at http://127.0.0.1:11434), python3 (only for cloning your own voice) and the voice models. None of them is required for the command line.

What the commands cost#

Measured by packages/tools/dev/benchmark.mjs, which starts each command as a fresh process and records its time and its largest use of memory. The same script fails a build when a command goes past its budget. The numbers below are copied from its output.

Command	Cold start	Peak memory	Budget (time and memory)
construct --version	50 ms	45 MB	0.6 s and 120 MB
construct validate (small project)	480 ms	135 MB	3 s and 400 MB
construct summarize (small project)	460 ms	130 MB	3 s and 400 MB
construct decide --requirement	190 ms	75 MB	0.6 s and 150 MB
Cockpit server, start until it answers	320 ms	95 MB	10 s and 400 MB

The small project is the architecture-valid-react-spa fixture, so a large project takes longer; the numbers show what starting Construct costs, not what your project costs.