| Version | Supported |
|---|---|
| 0.1.x | ✅ |
Please do not open a public issue for security vulnerabilities. Instead, open a private security advisory or contact the maintainers directly. We aim to acknowledge reports within 72 hours.
Orchevia executes local commands on your behalf. The design principles:
- Explicit execution only. Opening, importing, or previewing a workflow never runs anything. Runs start from the ▶ Run button or an approved resume.
- No implicit shell. Executables and argument lists are passed separately to the OS; shell metacharacters in arguments are inert. Command previews are shell-escaped for display only. Command nodes can explicitly invoke a shell or interpreter.
- Visible commands. The inspector shows exactly which executable + arguments will run; run history records exit codes, stdout/stderr (truncated), and errors.
- Cancellation + timeouts. Runs can be cancelled; in-flight children and, on Linux and macOS, the processes in their process group are stopped. Commands have a bounded timeout.
- Authentication stays in your tools. Orchevia has no credential fields and does not persist environment dumps. Workflow text and command output are stored locally and may contain sensitive information. Do not paste credentials into workflow fields; review output before sharing.
- Untrusted content. Ticket/PR text, imported workflows, and AI output are treated as untrusted input — review before running.
- Local-first. No analytics, no telemetry, no accounts, no Orchevia cloud. Third-party CLIs (Claude Code, Codex) may contact their own providers per their own behavior; that is outside Orchevia's control and is disclosed in the Environment screen.
- macOS builds are ad-hoc signed but not notarized, so the first launch needs Open Anyway (documented in README). Each release refuses an app that is not signed as a whole bundle and launches the DMG's copy before publishing.
- Approval gates pause durably, but a run interrupted by a crash mid-command is marked failed on next launch rather than resumed mid-command — resuming arbitrary external processes automatically would be unsafe.