AIGRaaS
Product

Custom Rulesets

Domain-specific guardrail configurations, no code required

Describe your domain in plain English. AIGRaaS generates a complete ruleset with decision trees, harm variables, and honesty checks calibrated to your use case.

4

Decision trees

8

Harm variables

7

Honesty components

Try in PlaygroundHow it Works

Mega-prompts don't scale. Every rule you bolt onto a system prompt makes it slower, more fragile, and easier for adversarial users to jailbreak. A ruleset is the opposite: declarative, deterministic, and versioned — your guardrails get tested like code, not tuned like prose.

Declarative by design

A ruleset is a single JSON document that defines your decision trees, harm variables, honesty checks, and compliance modules. No code, no prompts — just a contract your AI cannot break.

healthcare-ruleset.json
{
  "name": "healthcare-intake-v1",
  "mode": "pre-delivery",
  "compliance": ["hipaa", "hitech"],
  "harm": {
    "medical_advice": { "block": true, "severity": "high" },
    "phi_disclosure": { "block": true, "severity": "critical" },
    "diagnosis": { "block": true, "severity": "critical" }
  },
  "honesty": {
    "calibration": true,
    "non_deceptive": true,
    "forthright": "when_safe"
  },
  "tests": ["thousand_users", "senior_employee", "dual_newspaper"]
}

Plain-English generator

Describe your domain in a paragraph. Get a complete ruleset with decision trees and harm thresholds calibrated to your use case.

Versioned and testable

Rulesets live in your repo alongside your code. Diff them, review them, roll them back — they're infrastructure, not prompts.

Override without forking

Layer custom rules on top of pre-built compliance modules. The principal hierarchy keeps platform > operator > user overrides clean.

Validated against three tests

Every ruleset is continuously validated against the 1,000 Users Test, the Senior Employee Test, and the Dual Newspaper Test.

A ruleset is never a prompt. It compiles to a deterministic decision tree the evaluation engine walks in microseconds.

0

LLMs running at ruleset evaluation time

Questions we get

Ready to try it?

Evaluate any AI response in the playground — no signup required.

Try in PlaygroundView Pricing