A local-first, AI-assisted desktop development environment (Electron + React + Monaco). Its workflow is Research → Understand → Plan → Review → Execute → Verify, with the human approving every consequential step.
Meta AI models are first-class (muse-spark-1.3 default,
muse-wonder-1.0 optional) via MODEL_API_KEY. The
provider registry supports adding more providers through
the documented contract, but no second live provider ships
yet. The app never silently switches providers or models.
Only inside AI prompts you approve — and only the context attached to that request. Indexing, search, planning, checkpoints, and history all run locally. There is no telemetry and no background uploading. See PRIVACY.md.
In your environment (.env in development, user env in
production). Never in settings files, logs, session
history, or diagnostics. Settings screens show presence
("key missing"), never values.
Windows is validated. macOS and Linux packaging is configured but runtime testing is pending — see PLATFORM_SUPPORT.md for the live matrix.
Treat them like any third-party code: only install extensions you trust. Each active extension runs in a dedicated child process with timeouts and memory caps behind explicit permission grants, but this is not a hardened hostile-code sandbox. Installs land disabled; nothing activates without you. See THREAT_MODEL.md.
PLATFORM → Updates checks a release feed on demand (none is configured yet, so checks report unavailable). When a feed exists, downloads are checksum-verified and staged; you install the staged file yourself. Nothing auto-installs, and failures leave your install untouched.
Bugs: GitHub Issues with the bug template (attach a redacted diagnostics export). Vulnerabilities: private report via GitHub Security Advisories only — never a public issue. See SECURITY.md.
Yes — start with DEVELOPMENT_SETUP.md and CONTRIBUTING.md. Tests, docs, and a security paragraph for privileged code are required.