Getting started
Introduction
AgentID is an AI governance and data protection platform that sits between your people, your software and AI providers.
AgentID gives security, compliance and engineering teams one place to control how AI is used. Every prompt that passes through an AgentID integration is checked against your policy before it reaches the AI provider.
What AgentID does
- Detects personal data and secrets in prompts, pasted text, files and tool results: names, e-mail addresses, phone numbers, addresses, national IDs, bank details, API keys, passwords and more.
- Masks what it finds with typed placeholders such as
<PERSON_NAME_1>or<EMAIL_1>, so the AI provider never receives the original value. - Blocks prompts that break a guardrail: prompt injection, toxic content, personal data or secrets you decided to block, and your own content policies.
- Restores the original values for the user, on their device or in your application's response, when you allow it. The provider and AgentID's logs keep the placeholders.
- Logs every check and response in an activity log, with encrypted storage of originals and audited access.
- Documents AI systems for governance: transparency, user guidance, risk, data governance, evaluations, incidents and technical documentation.
Where it works
| Integration | Protects | Setup |
|---|---|---|
| Browser extension | ChatGPT, Claude, Gemini and Microsoft Copilot in Chrome or Edge; discovers other AI web apps | Install, then connect with System ID and API key or an enrollment token |
| Claude plugin | Claude Cowork and Claude Code (macOS, Windows, Linux) | Upload a ZIP in Claude settings |
| Claude Desktop companion | Ordinary chats in Claude Desktop on Windows and macOS | One install command per computer |
| SDKs and API | LLM calls in your own backend, API or agent | npm install agentid-sdk, pip install agentid-sdk, or HTTP |
Who uses which part
- Security and IT register AI systems, choose guardrails, roll out the browser extension or Claude integration, and review blocked and masked prompts.
- Developers wrap their LLM calls with an SDK or call the Guard and Ingest APIs.
- Compliance teams complete transparency, user guidance and documentation for each AI system.