FAQ

Strazh is a safety gate for coding agents: a Tauri v2 desktop app (Rust + React + TypeScript) wrapping a local Python core that sits between you and any coding agent you dispatch (Claude Code, Codex, Cursor, or Devin). Every dispatch is checked against an allow-listed working directory before anything runs, with the environment scrubbed and the process tree cleanly killed on stop.

No. Strazh itself only runs as a local desktop app on your machine: there is no hosted version, and dispatched agents run as real local processes under your control. The only cloud-side component is strazh.graphenlabs.com, a separate Next.js marketing and account/license portal used for checkout and license management; it never sees what you dispatch or what an agent outputs.

Free is a permanent tier, capped at 5 dispatches per day, hard. Every core safety control, the allow-list gate, env scrub, tree-kill, durable audit trail, and all four agent adapters, is included at no cost, on Free.

No. Strazh dispatches to whichever coding agent CLI or API you already have configured (Claude Code, Codex, Cursor, Devin), without running inference itself. No local GPU or model runtime is required by Strazh's own gate.

Yes, for the CLI-backed adapters: Strazh resolves and dispatches to whatever's already on your PATH (Claude Code, Codex, Cursor) or configured via API key (Devin). It's the gate in front of the agents you already have installed.

Windows is the confirmed, tested path today. Strazh ships as a signed MSI and an NSIS installer, built and verified through a Windows-first packaging track. macOS and Linux are not currently packaged or verified.

Free is a hard 5 dispatches/day cap. Individual Pro raises that to 35/day plus $0.01-per-dispatch metered overage beyond it: same gate, higher ceiling. Enterprise is uncapped dispatch, sold sales-assisted rather than self-serve.

Yes: Free is designed as a permanent tier, not a trial. No safety control is withheld to force an upgrade; the tiers differ only on daily dispatch volume and how it's sold (self-serve vs. sales-assisted), never on whether the gate itself works.

Billing runs through Dodo Payments, not Stripe. Stripe doesn't onboard India-incorporated businesses as merchants of record, so Dodo handles checkout and payment processing for Individual Pro. Authentication (sign-in/sign-up) is handled separately by Clerk and is not coupled to billing.

Refund policy specifics aren't published here yet. Check back on our Terms page once it's live for the current policy.

A server-side allow-list check against your configured roots, resolved and verified before any process is spawned, including traversal-escape attempts (../ paths that resolve outside the allow-list). If the check fails, nothing runs; the rejection is logged with the reason and a suggested next step.

No, the dispatched agent has real filesystem access within the allow-listed root, by design, since it's meant to edit real repos. The allow-list is the control that matters: keep it scoped deliberately to the repos you actually want touched.

It's durably logged: every rejection, not just approved runs, gets an audit entry with the reason (directory not allowed, cap exceeded, no agent available) and, where applicable, a concrete next step like adding the directory to your allow list.

Strazh can act as an MCP server, letting external MCP clients (Claude Desktop, Cursor, Codex CLI) trigger a dispatch directly through the same gate, free on every tier, including Free. It's private in the sense that Strazh itself stays local and only responds to the MCP client you've explicitly connected; it does not phone home or share dispatch content with Graphenlabs.

License tokens are signed with Ed25519 by the account portal at purchase time and verified entirely offline by the desktop app against a public key baked into the build. There's no runtime check-in to a license server required for Strazh to know what tier you're on.