veto-agents 0.0.17

AI agents that pay for things on your behalf, with the safety built in. Hermes-core, governed by Veto.

Veto Agents

AI agents that pay for things, on your behalf, with the safety built in.

A curated set of consumer AI agents — each one designed from day zero to spend money to do real work for you, every action governed by Veto, every spend signed, every verdict verifiable.

Install

curl -fsSL https://raw.githubusercontent.com/veto-protocol/veto-agents/main/install.sh | bash

Then:

veto-agents                           # walks first-time setup (email, wallet, QR funding)
veto-agents install media             # add your first agent + paste the Replicate token it asks for
veto-agents media "make a neon jellyfish in cyberpunk rain"

Three commands. The agent generates the image, Veto authorizes the spend, you get a signed receipt at veto-ai.com/r/<uuid>. That's the loop.

Other install paths

• Already have pipx: pipx install veto-agents
• Already have Python + want to be reckless: pip install veto-agents (not recommended — pollutes global)
• Manual: clone the repo, pip install -e . in a venv

The bet

Agents are about to spend a lot of money. Today's general agent frameworks (Hermes, OpenClaw, n8n) treat payments as an afterthought — a tool the agent might call if you wire it up. Veto Agents inverts that: every agent in this catalog is built around the assumption that it has money and will spend it. Veto governance is not a feature — it's the only way the agent works at all.

How every agent behaves

Five non-negotiable principles every Veto Agent inherits — see PRINCIPLES.md for the full version.

1. Plan-then-execute. Show the plan + cost estimate first. Wait for explicit consent. Never auto-spend.
2. Cost transparency at every step. Show actuals as they happen, with the receipt URL inline.
3. Receipts for everything spendable. Every paid action produces a Veto-signed verdict at veto-ai.com/r/<uuid>.
4. Veto is the only spend gate. Every paid call authorizes through Veto, every time, no caching.
5. Always offer cheaper alternatives when they exist. Cost-conscious by default.

That predictability is the product. Every other consumer agent in 2026 is "agent just goes." Veto Agents is the agent that asks first.

What's in the box

Four agents, each Hermes-core with Veto governance preconfigured:

• Media — generates images, video, and audio for you. Pays Replicate / Runway / ElevenLabs per call. Headline agent.
• Build — deploys your code on the cheapest infra it can find. Pays Vercel / Modal / Replicate for compute. Dev headline.
• Research — does deep research using paid search and content. Pays Exa / Tavily / x402-gated sources.
• Inbox — handles email, calendar, and scheduling using paid AI and scheduling tools.

Each agent ships with a default Veto policy (caps, allowlists, intent rules), a wallet provisioned via Privy on first run, and a receipts feed showing every action it took and why.

How this connects to Veto

Veto already ships the trust substrate:

• Engine — 8-stage policy + risk evaluation
• Receipts — Ed25519-signed verdicts at veto-ai.com/r/<uuid>
• APPS — open policy schema
• VetoGuardedAccount — on-chain hard-stop contract

Veto Agents is the consumer surface that surfaces all of that. The agents are the front door; Veto is the load-bearing wall behind them. Same primitives, packaged for a non-developer to install and use.

See ARCHITECTURE.md for how the layers compose.

Status

v0 in design. Build sequence:

1. Media — weeks 1–6, ship first
2. Build — weeks 7–10
3. Research + Inbox — weeks 11–14

License

MIT. Each agent is a forkable template. Self-host on your own machine using vanilla Hermes, or run via the hosted Veto Agents PWA at app.veto-ai.com.

Where the credit goes

• Hermes Agent — Nous Research. The core runtime every agent runs on.
• Privy — embedded wallet provisioning so users never see a seed phrase.
• Veto Protocol — the governance, receipts, and on-chain enforcement layer.

---

Veto governs. The rail executes. The agent works.