Your AI Agent is one
bad decision away
from a disaster.
VantaGate is the Human-in-the-Loop API layer that intercepts high-stakes AI actions, routes them to a human via Slack or Email, and delivers a cryptographically-signed decision back to your agent — in one API call.
Setup in 5 minutes. No custom webhooks. No state machines. No weekend incidents.
The Problem
Deploying AI to production is a
liability problem, not a tech problem.
You've built the agent. It works in staging. But the moment it touches real money, real data, or real users — you're exposed.
Your AI will make a catastrophic mistake.
Not "might". Will. The only question is whether you have a circuit breaker in place when it does. Most teams don't — until after the incident report.
Building your own oversight is a 6-week detour.
Custom Slack webhooks, state machines, retry logic, audit tables — all just to pause your agent. Meanwhile your roadmap waits. There's a better way.
Your CISO is blocking production deploy.
"Where does the data go?" is a fair question. Without provable Zero-Retention and immutable audit logs, enterprise deals die in legal. Every time.
The Solution
A 3-step protocol. One API call.
Zero infrastructure debt.
VantaGate sits between your agent and the real world. It never slows your agent down — it parks it safely while humans decide.
Agent Pauses
Agent hits a decision point. Calls POST /checkpoint.
Your agent serializes the context as JSON and fires one API call. VantaGate encrypts it, queues notifications, and responds PENDING in under 100ms. Your agent parks — no thread blocking, no I/O wait.
Human Decides
Approver gets a Slack message or Email with a secure one-time link.
They open the VantaGate decision UI — a clean, mobile-optimized dashboard. They review the full payload, see context, and click Approve or Reject with a mandatory reason. The payload is then permanently purged.
Agent Resumes
Signed decision delivered to your callback_url or via polling.
VantaGate fires an HTTPS POST to your callback URL, signed with HMAC-SHA256. Verify the signature, read the decision, resume or halt. No public URL? Use polling — both patterns are first-class.
The approver decision UI. Mobile-optimized. No login required — secure one-time magic link.
Objection Destroyed
"Won't adding a human approval
layer slow my agent down?"
No. Here's why: VantaGate is fully async. Your agent fires one POST, gets a 202 Accepted in under 100ms, and parks. The human approval happens out-of-band. Your agent is never blocked on network I/O.
- Zero payload loss — encrypted at write, durable until decision. No message dropped under load.
- Polling or webhook — run behind a firewall or in a Lambda with no public URL? Use polling. Both patterns work equally well.
- 5 webhook retries — exponential back-off starting at 60s. Your callback_url can go down briefly and still receive the decision.
Latency breakdown
Your agent's execution time is dominated by the human's response window — not VantaGate's infrastructure.
Integration
From zero to protected agent
in 5 minutes flat.
No SDK. No wrapper library. No config files. Just one HTTP call — from any language, any runtime, any framework.
curl -X POST https://api.vanta-gate.com/v1/checkpoint \
-H "X-API-KEY: vg_live_YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"title": "Wire Transfer: $12,500 to vendor_id_99x",
"description": "Agent flagged: amount exceeds auto-approve threshold.",
"payload": {
"vendor_id": "vendor_99x",
"amount": 12500.00,
"currency": "USD",
"initiated_by": "billing-agent-v2"
},
"slack_webhook_url": "https://hooks.slack.com/services/T00/B00/XXXX",
"notify_email": "[email protected]",
"callback_url": "https://your-agent.com/webhook/vanta",
"timeout": "4h"
}'
# ← Agent pauses here. CFO gets a Slack message in < 1 second.
# ← VantaGate responds 202 in < 100ms. Zero blocking I/O.1
API call to pause your agent
<100ms
time to first response
0
custom webhooks to build
Enterprise Security
Built for the CISO conversation.
Every architectural decision in VantaGate starts with one question: "What's the minimal data we need to hold — and for how long?" The answer is always: as little as possible, for as short as possible.
Zero-Retention Purge Protocol
Payload encrypted AES-256 at rest. The moment a human decides, data is atomically overwritten with NULL — not "deleted", purged. We cannot reconstruct it. Neither can an attacker.
Immutable Audit Trail
Every decision produces an HMAC-SHA256 chain: payload hash, decision hash, decider identity (masked). Cryptographic proof your auditor can independently verify.
< 100ms Response. Zero Blocking.
VantaGate returns 202 Accepted in under 100ms. Notifications fire async via BullMQ. Your agent never waits for Slack I/O — it parks and polls, or receives a callback.
Stateless Slack. No OAuth Storage.
Your Slack webhook URL is sent per-request and purged with the payload. We never store it server-side. Your security model, your control.
Email Fallback — Always-On Delivery
If Slack is unavailable, approval emails fire automatically. Every checkpoint can have both channels. No approval goes undelivered.
MCP Server — Native Agent Integration
Published on npm and indexed in the Official Anthropic MCP Registry. Drop our MCP server into any Claude-based agent with a single config line.
Who It's For
Built for every team that ships AI.
Rapid Prototyping & MVPs
One rogue AI action can bankrupt you or torch your reputation overnight. VantaGate is your circuit breaker — 5 minutes to set up, 24/7 peace of mind. No custom webhook engineering required.
Scaling to Production
Investors and early clients want to see you're in control of your AI. VantaGate makes you look enterprise-ready on day one — immutable audit logs, compliance-friendly architecture, no extra headcount.
Client Deployments
Stop reinventing safety rails for every client project. One VantaGate account, every agent you build. Offer "human oversight" as a premium deliverable — powered by an API call that takes 5 minutes per project.
Compliance & SOC2 Workflows
Your CISO needs answers before sign-off. Give them: Zero-Retention (Purge Protocol), AES-256 encryption, HMAC-SHA256 audit trails, GDPR/SOX readiness. Every checkbox, provably met.
@vantagate/mcp-server
Official MCP Server
Native Claude Agent integration.
Already on npm. Already in the Anthropic Registry.
Running a Claude-based agent with the Model Context Protocol? VantaGate is available as a drop-in MCP server. Your agent gets a create_checkpoint tool natively — no HTTP boilerplate required.
{
"mcpServers": {
"vantagate": {
"command": "npx",
"args": ["-y", "@vantagate/mcp-server"],
"env": {
"VANTA_API_KEY": "YOUR_KEY"
}
}
}
}Pricing
Start free. Scale when you're ready.
Hard limits. No surprise overage charges. Ever. Start for free — no credit card required.
Free
Validate the integration
- 50 checkpoints / mo
- Email notifications
- Slack notifications
- 7-day log retention
- Max timeout: 24h
- Webhook callbacks
Pro
Ship with confidence
- 2,500 checkpoints / mo
- Email notifications
- Slack notifications
- 90-day log retention
- Max timeout: 7 days
- Webhook callbacks
Scale
Enterprise-grade control
- 25,000 checkpoints / mo
- Email notifications
- Slack notifications
- 365-day log retention
- Max timeout: 30 days
- Webhook callbacks
No contracts. Cancel anytime. Usage resets each month.
Your next agent deployment
should not be a leap of faith.
Add a human gate to every high-stakes action. Ship AI that you — and your clients — can actually trust.