Stop your AI agent
before it does something wrong.
Put a human reviewer between your AI agents and real-world actions - before they process refunds, send emails, or modify records. Integrates in minutes. Built-in compliance for EU AI Act, DORA, and GDPR.
Chatbots answer questions. AI agents execute actions - refunds, emails, database writes, deploys. Cheqpoint gates the risk.
No credit card required
✓ 14-day free trial · ✓ No credit card required · ✓ Cancel anytime
EU AI Act · DORA · GDPR · FCA · EBA compliant · SOC 2 ready · HIPAA ready
Drops in front of
Process Refund
Issue $340 refund · Card ••• 4421 · James Wilson
Cheqpoint AI: Duplicate charge detected within 2 hours. High-value refund - human sign-off required.
Live in your inbox · Slack · Mobile · API
Works with 100s of tools your team already uses
From AI Agent to human decision in seconds
From your AI Agent firing an action to a human decision in seconds.
The actions your AI shouldn't take alone
From refunds to deploys, any consequential AI action — across any team — can be paused and reviewed before it executes.
Process customer refund
$149 refund — new account, no delivery scan, first request.
Drop database table
DROP TABLE user_sessions — 2.1M rows. Production. Irreversible.
Send bulk email
40,000-subscriber campaign with an unverified tracking link.
Reverse transaction
$2,400 reversal — AI flagged a duplicate but original receipt is missing.
Update entitlements
Adding enterprise features mid-contract with no signed amendment.
Accept contract terms
Unlimited liability clause in section 8.3 — legal sign-off required.
When an AI acts alone, your company owns the outcome
In Moffatt v. Air Canada (2024), Canada's Civil Resolution Tribunal held the airline liable for a refund policy its chatbot invented - rejecting the argument that the AI was a separate entity. Separately, a single prompt injection can turn a helpful agent into one that issues refunds, emails customers, or writes to your database on an attacker's behalf.
A human checkpoint on consequential actions is the control that keeps an autonomous mistake - or an exploit - from becoming a liability.
Built for every team running AI
Real-world scenarios where Cheqpoint provides a safety net. Try approving, declining, or modifying a request below.
Your support AI handles hundreds of requests per day. Any refund over $50 or account change that affects billing must pause for human approval before it executes.
AI Reasoning
“Tracking shows no delivery. First refund request.”
The approval inbox
A fast, simple inbox for your team. Zero technical knowledge required.
See the approval flow in action
Submit a request, review it in the inbox, and watch the audit trail update in real time.
Want a personal walkthrough? Email us →
Works with any AI Agent
Cheqpoint adds a human checkpoint to Intercom Fin, Zendesk AI, ElevenLabs, or any custom AI Agent - before it acts.
CS Platforms
Drop in alongside Intercom Fin, Zendesk AI, Freshdesk, and ElevenLabs. Trigger via webhook or use our native integrations - no complex code changes needed.
Custom AI Agents
Protect AI Agent actions with a single SDK call. Supports Python, Node.js, OpenAI, Anthropic, and the Vercel AI SDK.
No-Code Automations
Connect via Zapier or n8n. No SDK needed. Any automation that hits a webhook can route through Cheqpoint for human approval.
Don't spend a quarter building an approval queue
An approval layer is months of undifferentiated plumbing - review queues, Slack integrations, SLA timers, escalation, audit trails. Or one line of code.
Build it in-house
~3 months- Approval queue UI + real-time inbox
- Slack / Teams / email notifications
- SLA timers, escalation & business-hours logic
- Immutable audit trail + CSV export
- Roles, watchers & out-of-office delegation
- On-call rotations and per-assistant kill switches
Wrap one line with Cheqpoint
~5 min// Runs the refund only if a human approves
await cheqpoint.guard(
"Process refund",
{ orderId, amount },
() => stripe.refunds.create({ ... }),
);- Inbox, Slack buttons & audit trail included
- SLA, escalation & working hours built in
- Fail-safe: the action can't fire without a human
Already using Intercom or Zendesk?
Their AI Assistants resolve customer requests and issue refunds autonomously. Cheqpoint adds a human-in-the-loop approval layer before high-stakes actions execute.
Without Cheqpoint
1. Intercom Fin AI receives request
2. AI decides: issue $200 refund
3. Refund fires automatically
With Cheqpoint
1. Intercom Fin AI receives request
2. AI decides: issue $200 refund
cheq.checkpoint()one line of code3. Cheqpoint called
4. Human approves in 23 seconds
5. Refund fires safely
Works alongside Intercom, Zendesk, Freshdesk, Parloa and any platform with a webhook or API.
See integrations →Human judgment where it matters.
Policy everywhere else.
Approving every action by hand doesn't scale - and rubber-stamping defeats the point. Cheqpoint routes the risky 5% to humans, auto-decides the rest by policy, and records everything for your auditors.
Auto-decision rules
Set thresholds on amount, risk level, and AI certainty. Low-risk actions clear instantly; only the consequential ones reach a reviewer. Test changes safely in the Rule Playground before they go live.
Kill switch per Agent
An agent misbehaving in production? Pause it in one click. Every pending action holds until you turn it back on - no deploy, no code change.
Audit trail & compliance export
Every request, rule match, and human decision is logged with who, what, and when. Filter by regulatory tag and export evidence in CSV for SOC 2, internal audit, or regulators.
Built for the EU AI Act
Article 14 requires effective human oversight of high-risk AI systems. Cheqpoint gives you the oversight mechanism and the documented proof it operated - on day one.
Business hours SLA pausing
Configure working days and hours so SLA timers pause overnight and on weekends. Requests submitted out of hours are batched into a morning digest - no missed deadlines, no inbox dread at 7am.
Out-of-office delegation
Reviewers can set an out-of-office with a return date and delegate incoming approvals to a colleague. Requests reroute automatically - no manual handoff, no approvals silently stalling.
< 20s
to review and decide
Approve or decline via Slack, email, or web - no context switching
3 lines
of code to integrate
Wrap any AI action in a checkpoint - no infrastructure changes needed
100%
of decisions audited
Who approved what, when, and why - exportable for regulators
How teams use Cheqpoint
Block a fraudulent refund before it lands
An AI Agent raises a suspicious $400 refund for review. The reviewer sees the duplicate charge, declines in one tap, and the audit trail captures why.
High-risk action intercepted
Integrated in under 5 minutes
Install the SDK, wrap one action with cheqpoint.checkpoint(), and your first real approval request appears in the inbox. No infrastructure changes.
< 5 min to first checkpoint
Audit-ready from day one
Every request, decision, and reason code is logged automatically with timestamps and reviewer identity - ready for SOC 2 and EU AI Act evidence requests.
Zero-config compliance trail
Simple, transparent pricing
Free plan forever. 14-day Growth trial included - no credit card required.
Frequently asked questions
Get product updates
New SDK releases, integration guides, and feature announcements from the Cheqpoint team. Unsubscribe anytime.