Home/veto ai

Search alias

Veto for AI agents: runtime authorization before tools execute.

People search “veto ai”, “vetoai”, “veto ia”, and “veto agent”, but the product is Veto. It is runtime action authorization for AI agents: policy checks before tools execute.

BLUF

Veto sits after the model proposes an action and before the tool runs. It checks the tool name, arguments, actor, tenant, environment, risk, and approval state, then returns allow, block, or require approval.

Quick start docs Book authorization review

What Veto authorizes

Veto authorizes actions, not prompts. The protected unit is the proposed tool call: the exact function or MCP tool, its arguments, the identity context, and the policy that should apply at that moment.

Tool name and declared capability

Arguments, amounts, recipients, paths, and record IDs

Actor, user delegation, tenant, and project

Environment, time, rate, and session history

Risk class and current approval state

Matched YAML policy and outcome reason

Where it sits in an AI stack

Keep your model provider, tools, identity provider, and application code. Veto wraps the action boundary. The model can still plan; the tool only executes after policy says it may.

Model proposes tool call
  -> Veto checks policy and runtime context
  -> allow | block | require approval
  -> underlying tool executes only on allow

Allow

The action matches policy and the underlying tool executes with normal return handling.

Block

The action violates policy, so execution stops and the agent receives a controlled denial.

Require approval

The action pauses until a human reviews the tool, arguments, context, and policy match.

How it differs from prompt guardrails and auth

Layer	Question answered	Failure if missing
Authentication	Who is acting?	Impersonation or unknown actor.
Prompt guardrails	What should the model try to avoid?	Prompt injection or model drift.
Veto	May this exact action run now?	Valid credentials execute unauthorized work.

Frameworks and systems Veto can sit in front of

Use Veto with agent frameworks and tool surfaces where the model calls functions, browser actions, MCP tools, deployment APIs, financial operations, or external messaging workflows.

Framework integrations

OpenAI, Claude, Vercel AI SDK, MCP, LangChain, browser automation, TypeScript, and Python routes.

Docs quick start

Install the SDK, wrap tools, and test the first policy decision path.

Use cases

Money movement, PHI, claims, deploys, browser automation, external messages, and data access.

Authorization vs auth

Clarifies the search alias and the difference between identity and runtime action authorization.

Related routes for precise queries

Agent authz

Developer shorthand for runtime agent authorization after identity is known.

AI agent authorization

The broader model for can versus may, identity, policies, enforcement, and audit trails.

AI agent permissions

Map tools, arguments, context, and approval gates into policy.

AI agent access control

Access control at the tool-call boundary, not only RBAC or OAuth scopes.

Veto for AI agents FAQ

Is “veto ai” a separate product?

No. “veto ai,” “vetoai,” and “veto agent” are search phrases people use when they mean Veto. The product is Veto: runtime action authorization for AI agents, not a separate AI product line.

What does Veto do for AI agents?

Veto wraps the tools an AI agent can call. When the model proposes a tool call, Veto evaluates the tool name, arguments, actor, tenant, environment, policy, and approval state before the underlying tool executes.

How is Veto different from prompt guardrails?

Prompt guardrails try to influence model behavior. Veto enforces action policy after the model proposes a tool call and before execution. The model can ignore or reinterpret a prompt. It cannot bypass a tool wrapper that refuses to run the underlying function.

How does Veto work with OpenAI, Claude, Vercel AI SDK, MCP, and LangChain?

Veto sits around tools exposed through systems such as OpenAI function calls, Claude tool_use, Vercel AI SDK tools, MCP servers, and LangChain tools. The framework still plans and calls tools; Veto authorizes each proposed call before forwarding it.

Review your agent action boundary.

Bring a tool list, a risky workflow, or a policy draft.

Book authorization review