Guardrails API — Safety checks for AI agents

Add safety guardrails to AI agents via API, not hardcoded if/else

Problem

When building AI agents, almost every app needs checks like 'don't delete data unless the user confirmed.' You either write if/else logic for each case or rely on a custom LLM prompt whose free-text answer is hard to parse reliably. A developer is building exactly this and seeking feedback, confirming it's a recognized repetitive chore.

Solution

A 'guardrails as API' service: developers define rules (e.g., 'delete requires confirmation'), and agents call the API before sensitive actions. Supports rule versioning, audit logs, and replay, returning structured allow/deny/needs-confirmation results.

Users

Indie developers and small teams building AI agents, especially those involving write operations, deletions, or payments

MVP

REST API that takes an action description and returns allow/deny/needs-confirmation; rule configuration UI (JSON or simple DSL); request audit log; rule versioning and replay; Python/TypeScript SDK

Revenue

Usage-based pricing (free tier + $0.5-2 per 1k calls after); team plan adds rule collaboration and audit export

Why now

AI agents are moving from experiments to production, making guardrails a must-have, yet no de facto standard exists; indie devs can define the interface and build a rule library fast