Dashboard
Integration targets
The integration target of an AI system decides how it connects to AgentID and which tabs and settings it offers.
You choose the integration target when you register a system.
Available targets
| Target | What it protects | Connects with | Guide |
|---|---|---|---|
| Browser Extension | Prompts, pasted text, files and answers in ChatGPT, Claude, Gemini and Microsoft Copilot; discovery of other AI web apps | System ID and API key, or an enrollment token | Browser extension |
| Claude Desktop | Cowork and Claude Code (plugin), ordinary Claude Desktop chats on Windows and macOS (companion) | A connected plugin ZIP and a per-computer install command; no API key | Claude Cowork and Code |
| SDK / API | LLM calls in your backend, API or agent runtime | API key | Node.js SDK, Guard API |
ChatGPT Desktop & Codex is announced as Coming soon.
Targets on existing systems
Some systems created earlier use targets that are no longer offered for new systems:
- Claude Desktop Chat: the Windows companion for Claude Desktop chat only. New systems use Claude Desktop, which covers chat, Cowork and Code.
- Hybrid: browser protection and SDK / API traffic under one policy.
What differs by target
| Feature | Browser Extension | Claude Desktop | SDK / API |
|---|---|---|---|
| API keys on the Integration tab | Yes, plus enrollment tokens | No (plugin and device activation instead) | Yes |
| Block Prompt Injection switch | No | No | Yes |
| Restore masked values for the end user | Yes | Yes | Yes |
| Prompt injection checks | No | No | Yes |
| AI Systems Discovery tab (admins) | Yes | No | No |
| Compliance tabs (transparency, risk, documentation, ...) | No | No | Yes |
Hybrid systems have the compliance tabs, the AI Systems Discovery tab and a Prompts tab. Personal data, secrets, toxicity and content policies are checked for every target.