Start here: from installation to your first local result

1. Install the application

Use Node.js >=22.13.0 and a currently supported native package. macOS Apple Silicon and Linux ARM64/x64 are available; Linux requires glibc >=2.39. On Windows, use the Linux x64 environment inside WSL2. Read the release evidence for actual platform testing.

Keep optional npm dependencies enabled. npm installs the app and matching native core; DFT and MD scientific solvers are installed in the next step. Docker, a global Python installation, a source checkout, and a Phynite login are not required for the published local app.

Example / commands

node --version
npm install -g phynite-lab@latest
phynite-lab

2. Recognize the workspace

Anonymous local workspace — Real local demo workspace. Start with New project; Cloud login is optional.

Open Workspace and choose New project. Name the study by the question or material, such as Silicon ground state.
Open the project. Canvas connects scientific modules; opening a module shows its editor. Keep one coherent study in a project and name material variants clearly.
Use Settings to configure this computer. A missing profile in an unsigned local workspace is expected; AI provider login and Phynite Cloud login are separate optional connections.

3. Prepare this computer for science

Scientific runtime setup in Settings — The current computer reports its own installed runtimes and isolation state. Your values may differ.

Open Settings → Compute. Confirm the detected operating system and architecture, available CPU/RAM, and the selected workspace location.
Install DFT for Quantum ESPRESSO, MD/analysis for LAMMPS and native analysis, or both. Wait for the actual installation status and capability checks; a download finishing alone does not prove the solver is ready.
Verify the isolation probe. If it is blocked, read the operating-system diagnostic and repair that prerequisite. Do not disable confinement or use an unverified runtime to bypass the message.
For DFT, install the explicitly selected SSSP Precision or Efficiency library in DFT pseudopotentials, or provide the exact UPF files required by the input. For MD, select a compatible force field and verify element mapping, license, and model source.
Set a modest CPU/RAM allocation and parallel-job count for the first exercise. Leave room for the operating system and temporary files. More cores do not guarantee faster execution.

4. Run one small, reviewable calculation

Import a known small structure, inspect species, atom count, cell vectors, coordinates, and units, and retain its original file. Use the silicon tutorial for DFT or the argon tutorial for MD.
Add the required module and connect its source. For a first DFT exercise, use SCF / Total energy; for MD, review one short Simple stage before adding a larger chain.
Inspect the exact generated input, method, pseudopotential or force field, sampling, and output settings. Presets are starting points, not evidence of scientific convergence.
Choose Your computer, then Run. In Review run, inspect the Compute plan and prerequisite list. Resolve blocked runtime, potential, isolation, input, or resource checks before Start run.
Follow Run history and open the module’s Execution result. Queued means waiting; Running means work has started; Completed means retained results are available. Open Execution log and inspect the numerical result before accepting it.

5. Save, reopen, and protect the result

Retain the exact input, original structure, solver/runtime identity, potential checksums, provenance, and all artifacts needed by downstream calculations.
Wait for the app’s save status. Finish or cancel active jobs, stop the launcher, and reopen the same data directory to verify the project and execution record remain available.
Export a project archive and keep an independent backup for important work. A browser draft, a figure image, and optional Cloud Sync each serve different purposes; none proves all scientific intermediate files were retained.

If this happens	Next action
Start is blocked or the attempt fails immediately	Read the first setup/preflight reason. No solver log is expected if no solver process started.
A running solver fails	Keep the input, log tail, error, and available artifacts. Check the first solver error, disk, memory, and physical model.
Project or runtime seems missing after restart	Check status, doctor, the actual port, and the original --data-dir before reinstalling or deleting anything.