Terms
The terms of using Sevra.
Last updated July 16, 2026
These terms govern your use of Sevra. By creating an account you agree to them.
The service
Sevra is a control layer for AI agents. For actions your integration submits before execution, it evaluates the request against configured controls and returns Allow, Require approval, or Block. A held action can attach a case file; the hold remains enforced if case-file creation or enrichment is unavailable.
Your account
You are responsible for your account, the agents you connect, the scopes you grant, and the safety and autonomy settings you configure. Keep your credentials secure and tell us promptly if you suspect they are compromised.
Autonomy and responsibility
You are responsible for routing relevant actions through the guard, honoring every verdict, and preventing a provider call from bypassing that check. Configured agent levels do not control an action your integration never submits to Sevra.
You remain responsible for the actions you authorize your agents and systems to take, and for reviewing the escalations Sevra hands you.
Sevra is a control layer that reduces risk by design. It is not a guarantee against every possible failure.
Acceptable use
Do not use Sevra to break the law, bypass another party's security, or harm third parties. Do not attempt to widen an agent's scope beyond what you are authorized to grant.
Fees
Current plan prices, protected-workflow allowances, included guarded-action volume, and included active-human seats are shown on the pricing page. Gate includes five seats and Scale includes fifteen. Each additional active-human seat is $50 per month. Paid-seat admission requires verified workspace billing authority and provider confirmation and fails closed when either is unavailable. During early access, guarded-action usage beyond the included amount does not create an automatic overage charge or bypass the gate. A browser redirect alone does not assign a paid plan.
Availability
Current self-serve plans do not publish an uptime SLA. If the guard cannot evaluate or persist a routed decision, it returns Block without an execution token.
Changes
The date above identifies this published version. A later published version will carry its own revised date.
Contact
Questions about these terms: hello@sevra.dev.