Nod sits in front of the handful of actions your agents shouldn't take alone — sending money, emailing a stranger, deleting data, posting publicly. One API call, and a real, qualified human on your own team has to say yes before it happens.
Get your API key → Read the docs
Free during beta · no credit card · live key in under a minute
curl -X POST https://nod.dev/v1/review \
-H "Authorization: Bearer $NOD_API_KEY" \
-d '{"action_type":"spend_money",
"agent_reasoning":"Renew the expiring domain, $38",
"risk_level":"medium"}'
Email in, API key out — instantly. No waitlist, no sales call, no approval queue to get into the product.
One POST /v1/review right before the risky step, with what it's about to do and why. Nod hands back a pending request.
Whoever's on review duty gets a one-tap Approve / Reject link by email. Your agent polls — or gets a webhook — and only proceeds once someone has actually said go.
Agents are confident even when they're wrong. A model will execute a bad plan with exactly as much certainty as a good one. Nod is the one place in the loop where confidence doesn't substitute for a human actually looking.
There's no anonymous crowd of strangers rubber-stamping your agents' actions. Reviewers are people you've personally added to your account, each with their own login and name attached to every decision — accountability you can actually point to.
Every request, every decision, who made it and when, is logged — not buried in a Slack thread you'll never find again. If something goes wrong six months from now, you can reconstruct exactly who approved what.
The worst version of this product pings you about everything until you start approving on
autopilot — which defeats the entire point. Nod is deliberately narrow: you decide which
action_types are worth a human, not every tool call your agent makes. Wire it in
front of the handful of things that would actually be bad to get wrong — spending money,
talking to a stranger, touching production, going public — and leave everything else to run on
its own. Every request arrives with the agent's own reasoning attached, so approving takes one
glance, not an investigation.
One API call to ask. One API call (or a webhook) to find out the answer. Risk levels, free-text context fields for the agent's reasoning, and per-agent API keys so you can scope exactly what each agent is allowed to ask for. The full integration guide is written so you can hand the page straight to your coding agent and it can wire Nod in itself — no human developer required for that part, at least.
pending until someone decides — your
agent should be built to wait or time out gracefully on its own schedule. An optional
expires_at field lets you flag a request's intended deadline for the reviewer today;
automatic expiry enforcement is on the roadmap.