Phynite Lab

A local computational materials-science workspace, released under GNU AGPL v3 only (AGPL-3.0-only). Source and build instructions are available at https://github.com/phynite-lab/phynite-lab. Commercial use is permitted under the license. Users interacting with a modified version over a network must be offered its corresponding source, as required by AGPL section 13.

npm install -g phynite-lab
phynite-lab

Requires Node.js 22.13 or later. Each published version ships matching native packages for macOS Apple Silicon (darwin-arm64), Linux ARM64 (linux-arm64) and Linux x64 (linux-x64, glibc >=2.39), including current Omarchy/Arch Linux. Windows is supported through WSL2, which uses the Linux x64 package and launcher. Linux qualification runs on matching release builders; direct testing on an arbitrary Omarchy desktop or Windows host is not claimed; consult the release evidence for the targets exercised for the installed version. npm selects the exact-version native platform optional package containing the production app, frozen local core and Linux sandbox; it does not require a global Python installation, AWS CLI, or Docker. Scientific solver runtimes are installed separately from the application's setup page.

On Omarchy, run the launcher as your normal desktop user. Its bundled bubblewrap needs unprivileged user namespaces; Setup verifies actual filesystem/network isolation before enabling scientific execution. A blocked probe displays the repair needed instead of weakening OS policy. Browser opening uses xdg-open; --no-open also prints the local URL for manual opening.

The application binds to 127.0.0.1, opens a browser, and keeps user data in ~/.phynite-lab. A busy default port is replaced with an available local port. Press Ctrl+C to stop the application without removing saved data.

phynite-lab --no-open --port 3000 --data-dir /path/to/workspace
phynite-lab status
phynite-lab stop
phynite-lab doctor

Use the same --data-dir for status, stop, and doctor when using a custom location. Repeating the launch command opens the existing instance for that data directory. No installer runs during npm installation. Keep npm optional dependencies enabled. Incomplete or mismatched distributions fail with an actionable error rather than falling back to development fixtures. distribution.json identifies the exact public source commit and source archive for each installed release. The application also provides a source link. Updated packages do not remove saved data:

npm install -g phynite-lab@latest

The launcher checks https://registry.npmjs.org/phynite-lab/latest without credentials. Successful metadata is cached for up to six hours and an unavailable registry response for five minutes. The update notice is informational only and never runs npm for you. Offline use remains available; stop the current instance, run the update command, and start it again. To verify the published metadata from a terminal, run npm view phynite-lab version and inspect the installed launcher with phynite-lab --version.

The first solver installation downloads the chosen runtime from the setup page.

Complete English and Indonesian user documentation is available at https://phynitelab.com/guide, including the npm install/update path, local AI provider accounts, Cloud Sync, storage and QE intermediate-state guidance, Settings, and troubleshooting. The guide links to the matching source and downloadable Markdown copies.

Use the visual task paths in the guide to move from first launch to a verified QE or MD run. Screenshots use a clean demo workspace and are reused in the English and Indonesian guides with translated captions.

First run checklist: open Settings → Compute, verify the platform, solver capability, and isolation probe, install DFT or MD/analysis, install the required pseudopotentials (or provide exact UPF inputs), choose Your computer in the run review, and run a small reference calculation. Inspect the actual queued, running, and final state, the first solver error, execution log, provenance, checksums, and numerical output. A preflight failure occurs before the solver starts and therefore has no solver output; a runtime failure has a solver log and any artifacts that were retained.

The default workspace is ~/.phynite-lab; pass --data-dir to use another location. SQLite is the metadata store and the filesystem holds project artifacts, runtime installations, and logs. The local Storage page measures only inventoried project artifacts. It excludes SQLite metadata, installed solvers, logs, and unrelated files, and local mode has no fixed 10 GB quota or automatic seven-day cleanup. Available disk is the practical limit, so export important projects and keep a separate backup.

QE intermediate state can dwarf the final output. disk_io, wavefunctions, and tmp/ files may be needed for bands, DOS, phonons, projected workflows, or external interoperability. Keep them until downstream workflows finish and retain the exact input, pseudopotential identity/checksum, provenance, and state artifacts. Wannier90 remains an external/manual workflow unless an explicit integration is documented: QE documentation, Wannier90 documentation.

Cloud storage is separate from this disk-backed workspace. Cloud accounts have their own quota, retention, and Trash behavior; signing in does not upload local projects, and Cloud Sync is enabled per project. Cloud compute uses a separate input-transfer review.

Cloud and AI connections are optional and configured inside the application. Local AI uses the official Codex or Antigravity CLI already installed on this computer. Open Settings → AI provider accounts and sign in; models connect and verify automatically. An existing CLI sign-in can be reused. Phynite never reads or copies CLI session files or provider tokens. The bridge runs on loopback with shell, MCP, plugin, browser, and network tools disabled; a model is unavailable until its text, streaming, and host-validated tool-proposal probe succeeds. A loopback OpenAI-compatible Antigravity uses a restricted profile, explicit permission denials and a native tool denial hook. A synthetic startup turn verifies the loaded guard before any project context is sent. Unsupported guard or protocol versions fail closed. Hosted Phynite AI is not used by the local application. Cloud sign-in requires a release assembled with public cloud connection metadata; an unconfigured release remains available for local work. Signing in does not enable Cloud Sync for a project. The launcher ignores inherited cloud settings. Provider usage refreshes automatically while the settings page is visible. Codex reports session and weekly percentages through its app-server; Antigravity 1.4.11 reports model-group limits through the read-only /usage command. Each window shows the provider's reset time and a countdown. These percentages are subscription limits, not token counts. Failed refreshes retain the last report with its timestamp; unsupported or missing reports are shown as unavailable. Third-party components retain their own licenses; see THIRD_PARTY_NOTICES.md in the assembled release.