AgentPing
Security for AI agents and AI chats

Your team uses AI every day. Your secrets stay home.

AgentPing sits on every computer and checks what coding agents and AI chats are about to send. API keys, passwords, customer data and ID numbers are replaced or blocked before they leave the machine, and your security team sees what happened.

macOS and Windows · Claude Code, Codex, Cursor, GitHub Copilot, ChatGPT, Claude, Gemini and more · Content never leaves the computer

Protects the AI tools your team already uses

Claude CodeOpenAI CodexCursorGitHub CopilotVS Code ChatGPTClaudeGeminiPerplexityGrokMCP servers
The problem

AI agents read everything. Then they send it to a model.

A coding agent opens .env to debug a connection. Someone pastes a customer list into a chat to summarise it. A screenshot with a key goes into a prompt. Each one copies data you're responsible for to a service you don't control, and nobody notices.

Keys and passwords

Cloud keys, tokens and database passwords leave in prompts, pasted logs and files agents open.

Customer data

Names, card numbers and national ID numbers end up in chats and model requests, beyond your data processing agreements.

No record

Security teams can't see which tools are used, what was sent or what each team spends on AI.

How it works

Checked on the computer, before anything is sent

No gateway to route traffic through and no content in our cloud. AgentPing runs where the work happens.

Install once

One installer per computer, or roll it out with your device management. It connects to Claude Code, Codex, Cursor, Copilot, MCP servers and the browser by itself.

Every send is checked

Prompts, files an agent opens, shell commands, uploads and pastes are inspected locally in milliseconds, with 259 detectors, document text extraction and text recognition for images.

Replace, block, record

Sensitive values become placeholders so the work carries on, or the send is blocked, as your policy says. Findings, never the values, go to your console.

What it catches

Built for what actually leaks

Secrets and credentials

API keys for every major cloud and SaaS, tokens, private keys, passwords in text and database connection strings. Checksums and context keep false alarms down.

Personal data and ID numbers

Card numbers, US SSN, Aadhaar, PAN, passports, driving licences and national identity numbers from over 40 countries, each validated by its own checksum.

Files and screenshots

Uploads and files agents open: text, CSV, PDF, Word, Excel, PowerPoint, archives and images. Scans and screenshots are read with on-device text recognition, and found values are blacked out.

Config and key files

.env, cloud credentials, SSH keys and production configs: agents are stopped from reading them, whether through a file tool or cat in a shell.

Agent mistakes

Packages that don't exist (a supply-chain risk), claims a session can't back up, and other errors agents make, caught as they happen.

Usage and cost

Which AI tools are used, by which team, on which project, and what each costs at list price, per project and per device.

Privacy by design

Your content never reaches us

Inspection happens on the computer. The console receives what was found, where and what happened, never the text, file or value itself.

  • Findings carry a description, location and a keyed fingerprint, not the secret
  • Previews are masked on the device (AKIA••••7Q2X)
  • Each workspace's data is isolated by the database itself (row-level security)
  • Protection keeps working offline with the last policy
  • Self-host the console in your own cloud for full control

How AgentPing handles data →

// What the console receives for a finding
{
  "rule": "aws_access_key_id",
  "description": "AWS access key ID",
  "classification": "restricted",
  "surface": "prompt",
  "path": "config/.env",
  "preview": "AKIA••••••••7Q2X",
  "fingerprint": "hmac-sha256:7c1e…",
  "outcome": "redacted before sending"
}
For security and IT teams

One console for every computer and every AI tool

Findings, projects and devices

See every finding with its severity and what happened to it, by project, by device and by person. Triage, mark false positives and export to your SIEM.

Policies your way

Replace or block, per type of data. Start in monitor mode, simulate a policy against last week's findings, then publish. Computers pick it up within minutes.

Enterprise sign-in

Single sign-on with Okta, Microsoft Entra ID, Google Workspace or any SAML/OIDC provider, user directory sync (SCIM), multi-factor sign-in. Admin, analyst and device-user roles.

Coverage and audit

Know which computers are protected, which agents are connected and which run an old policy. Every admin action is in the audit log.

Protect your first five seats today

Your first 5 seats are free, with every feature. Then $2.99 per seat per month. No credit card to start.