What is Veto?

Veto is a guardrail system for AI agent tool calls. It sits between an AI agent and its tools — intercepting every tool call, validating it against rules and policies, and allowing, blocking, or escalating for human approval.

The agent is unaware of the guardrail. The tool interface is preserved.

How it works

1. AI agent calls a tool (e.g. send_email({to: "...", body: "..."}))
2. Veto intercepts the call via the SDK
3. Arguments are validated against your rules (YAML constraints + LLM semantic checks)
4. The call is allowed, blocked, or escalated based on the result
5. If allowed, the tool executes normally. If blocked, the agent receives an error

┌──────────┐     ┌───────┐     ┌──────────┐
│ AI Agent │────▶│ Veto  │────▶│  Tools   │
│          │     │       │     │          │
│  calls   │     │ rules │     │ execute  │
│  tools   │     │ + LLM │     │ actions  │
└──────────┘     └───────┘     └──────────┘

Features

• Provider agnostic — works with OpenAI, Anthropic, Google, LangChain, Vercel AI SDK
• YAML rules — define constraints with simple, declarative YAML
• LLM validation — semantic checks for cases static rules can't cover
• Multiple validation modes — API (remote server), custom (direct LLM), or kernel (local Ollama)
• TypeScript + Python SDKs — first-class support for both languages
• Dashboard — real-time monitoring at runveto.com

Quick example

import { Veto } from 'veto-sdk';

const veto = await Veto.init();
const wrappedTools = veto.wrap(myTools);

// Pass wrapped tools to your agent — it works exactly the same,
// but every call is now validated against your rules
const agent = createAgent({ tools: wrappedTools });

Next steps

• Installation — install the SDK for TypeScript or Python
• Quick Start — get Veto running in 5 minutes
• Rule Format — learn the YAML rule syntax