Artificial Intelligence Access Guard

AI spend that can’t run away, even when your auth breaks

A checkpoint in front of every AI request, so a public chat, a hijacked account or a route nobody meant to expose can’t run up your bill.

AI Guard is a small service your server calls before each request to a model. One check() enforces spend caps, scores the request for abuse and tells you what to do: go ahead, slow down, use a cheaper model, ask the browser for a proof-of-work, or refuse. A record() afterwards reports what the call really cost.

It sits outside your app’s login, so its limits still hold when that login fails. And it fails open: if AI Guard is ever slow or unreachable, the answer is “go ahead”. Our outage never becomes yours.

The problem

Three ways AI features leak money

Public endpoints

No-login chats, demos and AI search. Per-address limits lose to a cheap proxy list: abusers spread the spend across thousands of addresses.

Hijacked and farmed accounts

A stolen session inherits its owner’s quota, and free tiers get farmed by the hundred. A limiter that trusts your login trusts the attacker too.

The route you didn’t mean to expose

Fast-built apps ship AI routes without authentication. AI Guard’s caps don’t depend on your login, so they still hold when it doesn’t.

How it works

One call before the model, one after

const d = await aiag.check({ user_id, ip: req.ip, ua: req.get('user-agent'),
                             model: 'gpt-4.1-mini', prompt_text: message });
if (d.action === 'block')     return res.status(429).end();
if (d.action === 'challenge') return res.json({ challenge: d.challenge });
// ...call the model, then report what it really cost:
await aiag.record(d.check_id, { tokens_in, tokens_out, model });

Every request gets a risk score, weighed against what is at stake: its cost times how fast that identity is going. Cheap requests at a human pace get slack; fast or expensive ones get scrutiny. People pass the soft steps; scripts pay for them.

Read the integration guide

Signals

What it scores

SignalWhat it catches
Spend capsPer end user, per feature and overall, checked before the call is made
VelocityOne identity, or one shared address, going faster than people type
Account farmsMany brand-new identities from one address, especially ones created minutes apart
Prompt repetitionThe same prompt replayed by one identity, as resellers and scripts do
TimingRequests spaced like a metronome instead of like a person
TrustIdentities earn headroom by behaving; new ones start small
Proof-of-workOn no-login pages, a single-use puzzle that costs scripts computing time on every request

Prompts aren’t stored: only a short fingerprint to spot repeats, plus the first 500 characters of flagged ones, kept for 30 days so you can see why they were flagged.

Dashboard

See every decision and why it was made

Each account has a private dashboard: checks, decisions, spend let through and refused, tokens, a breakdown by site or feature, the identities to watch, and every flagged request with the signals behind it in plain language.

Its link arrives with your API key and looks like /ai-guard-dashboard?key=aiagv_…. That key is view-only: anyone with the link can read your stats but can’t use your quota. Lost it? Request a new link with the same email.

Open your dashboard

Pricing

Free to start

Free

$0

100,000 checks a month: the checkpoint API and SDK, spend caps, abuse scoring and proof-of-work challenges.

Pro, coming soon

Key vault

Your provider keys live behind the checkpoint, so even a fully compromised app can’t spend past your policy. Higher volume and alerts.

Do your AI features answer strangers?

The free scanner looks for AI routes that reply without credentials, gates that only check a forgeable header, and provider keys exposed in your site’s code.

Scan your site free