What is Vettly?

Vettly is decision infrastructure for automated judgments that require defensibility.

At its core, Vettly is a system that:

• Accepts content and context
• Applies explicit, versioned policy rules
• Routes to one or more AI providers
• Returns a graduated decision (allow/warn/flag/block)
• Preserves an immutable record of what was decided, why, and under what policy version
• Provides mechanisms for appeal, review, and audit

The Problem with Classification APIs

Classification APIs (OpenAI Moderation, Perspective API, etc.) answer a simple question:

"What category does this content match?"

They return a label and a confidence score. That's it.

But classification is not governance. When regulators investigate, when lawyers subpoena, when users contest decisions, you need more than a score. You need:

• What was the exact content evaluated?
• What policy rules were in effect?
• What thresholds triggered the action?
• Who made the decision and when?
• Can we re-evaluate under current policy?

Classification APIs don't answer these questions. Vettly does.

Decision Infrastructure

Vettly converts automated judgments into auditable, appealable, reproducible decisions governed by explicit policy.

Every Decision Has an ID

const decision = await vettly.check({
  content: post.body,
  context: { userId: user.id }
})

// decision.id = "dec_abc123..."
// Use this ID for:
// - Appeals
// - Audit reports
// - Legal discovery
// - Compliance documentation

Policy is Versioned

# policy-v2.yaml
rules:
  - category: harassment
    threshold: 0.4
    action: flag

When you update your policy, Vettly tracks:

• Which version was active when each decision was made
• How decisions would differ under the new policy
• Complete history of policy changes

Decisions are Preserved

Every decision includes:

• Input hash: Verify the exact content evaluated
• Policy version: Know which rules applied
• Provider response: Raw scores and classifications
• Action taken: The final decision
• Context: User ID, session, timestamp, custom metadata

Appeals are Built In

When users contest decisions:

1. Retrieve the original decision by ID
2. Show what policy rules triggered the action
3. Re-evaluate under current policy if rules have changed
4. Preserve complete appeal trail

What Vettly Is Not

Not Just a Classifier

Classifiers produce labels. Vettly produces defensible decisions with audit trails.

Not a Feature—It's Infrastructure

Content moderation is a feature. Governance infrastructure is foundational. Vettly is the latter.

Not Vendor Lock-In

Vettly abstracts across providers (OpenAI, Gemini, Azure, Hive). Same policy, same decision semantics, regardless of underlying classifier.

Who Needs Decision Infrastructure?

Platforms with Legal Exposure

A single undefended decision can become a lawsuit. Vettly provides the audit trail legal teams require.

Regulated Industries

Fintech, healthcare, EdTech—regulators require documented, consistent, auditable decision processes.

DSA Compliance

The Digital Services Act requires transparency in content decisions. Vettly provides the documentation automatically.

Multi-Jurisdiction Operations

Different jurisdictions require different policies. Vettly enables jurisdiction-specific rules with unified infrastructure.

The Technical Foundation

Vettly maintains:

• Policy-as-code: Versioned YAML rules defining thresholds, actions, and behaviors
• Decision identity: Unique ID linking every judgment to policy, input, and action
• Evidence chains: Complete context preserved for each decision
• Graduated actions: Four-tier response system (allow/warn/flag/block)
• Appeal hooks: Built-in workflow for contested decisions
• Provider abstraction: Consistent semantics across any AI provider

Next Steps

Ready to implement decision infrastructure?

• Getting Started →
• Decision Records →
• Policy Versioning →
• Audit Trails →