AgentPing
Security and privacy

We built AgentPing so we never see your data

A tool that stops leaks shouldn't become one. Everything AgentPing inspects stays on the computer it came from.

Inspected on the device

Prompts, files, pastes, uploads and commands are checked by the AgentPing engine on the computer. Text, files and values are never uploaded to AgentPing, to your console or anywhere else.

Findings, not content

The console receives what was found, where and what happened: a detector name, a location, a masked preview and a keyed fingerprint (to spot the same secret twice). Not the secret itself.

Works offline

Policies are cached on each computer, so protection doesn't depend on reaching the service. Findings queue locally and upload when the computer is back online.

What leaves a computer

Each event sent to the console has a fixed, published format. Anything outside it is rejected by the service.

  • Which tool and integration (for example Claude Code hooks, browser extension)
  • Project folder name and session, to group activity
  • For each finding: detector, category, classification, confidence, file path and line
  • A masked preview (AKIA••••7Q2X) and a keyed fingerprint
  • What the policy decided and whether anything was sent
  • For model requests through the proxy: model name and token counts

Never sent

  • Prompts, chat messages or responses
  • File contents, uploads or screenshots
  • Secret values or personal data
  • Command output

How the service is protected

Workspace isolation

Every workspace's rows are separated by Postgres row-level security, enforced by the database for every query, not only by application code. AgentPing staff see usage counts across workspaces, never findings.

Sign-in

People sign in through WorkOS: single sign-on with your identity provider, multi-factor sign-in, and directory sync to remove leavers automatically. Sessions are server-side and end on sign-out, role change or removal.

Devices and keys

Each computer signs in as a person and gets its own revocable credentials. Tokens and API keys are stored only as hashes, and computers can be signed out from the console at once.

Signed software

Releases are signed, and agentping update refuses anything whose signature doesn't match. The browser extension only talks to the engine on the same computer.

Audit trail

Every change to policies, people, devices and keys is in the workspace's audit log, including any action by AgentPing staff.

Your own cloud

With Enterprise self-hosted, the console, API and database run in your infrastructure, under your network rules, retention and backups.

Data retention

Events and findings are kept for 365 days by default. Each workspace can keep less, and deleting a workspace deletes its data.

Service providers

WorkOS handles sign-in (names and email addresses of console users and device users). Our hosting provider runs the service and its database. Neither receives inspected content, because the service never has it.

Report a vulnerability

Found a security problem in AgentPing? Email security@agentping.dev with the details and steps to reproduce. We acknowledge reports within two working days and keep you updated until it's fixed.

Please don't test against other customers' data, and give us time to fix the problem before sharing it publicly.

Email the security team