Using Sonomos with AI Platforms

Sonomos is designed specifically for the workflow of pasting or typing sensitive information into AI platforms. This guide walks through how to use it effectively on each major platform, and what to expect on platforms that aren’t fully covered yet.

Prototype platform coverage

The browser extension fully supports a small, focused list of platforms today (Claude.ai, Gemini, Grok). Every other site falls back to Send Guard warnings. Native AI desktop apps, IDE assistants, and local LLMs are explicit launch targets for Sonomos Desktop — they’re not reachable from a browser extension.

Before you start

1. Install Sonomos and verify the risk widget appears.
2. Enable Cloak for the platforms you use regularly.
3. Skim the Quickstart to internalize the traffic-light system.
4. Decide your default Send Guard mode (Banner + Modal is a good baseline — see Send Guard).

Coverage at a glance

Platform	Protection	Notes
Claude.ai (web)	Full automatic Cloak masking	Recommended platform for testing the flow
Google Gemini (web)	Full automatic Cloak masking	—
Grok on x.com	Full automatic Cloak masking	—
ChatGPT (web)	Send Guard fallback	Cloak integration in development
Perplexity	Send Guard fallback	—
Microsoft Copilot (web)	Send Guard fallback	—
Self-hosted UIs	Send Guard fallback	Any site with a normal <textarea>
Native AI apps	Not supported in extension	Coming in Sonomos Desktop
IDE AI assistants	Not supported in extension	Coming in Sonomos Desktop
Local LLMs	Not supported in extension	Coming in Sonomos Desktop

Platform-specific guidance

Claude.ai

Sonomos provides full automatic masking on Claude.ai. With Cloak enabled, PII in your messages is redacted before reaching Anthropic’s servers. The AI receives masked text, and your local view is restored with the original values.

Tips:

• The first message in a new conversation is the most important — that’s when the system establishes context. Make sure Cloak is on (look for the lock on the widget) before you send.
• If you’re sharing a file or pasting a long document, give Sonomos a second to finish scanning before submitting. The widget will turn from green to its final color when scanning is complete.

Google Gemini

Full automatic masking support. Sonomos intercepts and masks PII within your messages before they’re transmitted to Google.

Tips:

• Gemini’s “Drag in a file” interaction is covered. Dropped images are OCR’d and dropped PDFs are parsed.
• Workspace-tied Gemini (inside Docs / Sheets) is treated as the host app, not the standalone Gemini surface — Send Guard handles those entry points.

Grok (X / Twitter)

Full automatic masking support. Works the same as Claude.ai — outgoing messages are scanned and masked transparently.

Tips:

• Grok prompts entered from inside a tweet draft surface use a different composer than the dedicated Grok page. Both are covered, but the dedicated page is the cleanest test surface.

ChatGPT and others

These platforms are covered by Send Guard. You’ll receive a warning banner or blocking modal when PII is detected, but automatic masking is not yet available for these platforms.

Tip

If you primarily use ChatGPT, consider manually replacing PII with placeholders (e.g. [PATIENT], [SSN]) before submitting. Send Guard will catch what you miss, but manual replacement plus Sonomos is the most secure workflow for platforms without automatic masking.

Self-hosted / on-prem AI tools

Dagger detection works on any webpage regardless of the platform. Cloak’s automatic masking targets specific platforms, but Send Guard activates on any site where PII is detected in form fields. That means an internal “ChatGPT clone” on your network gets banner + modal warnings out of the box.

If you have a high-volume internal AI surface you’d like fully Cloak-supported, get in touch — those are exactly the integrations that inform the Sonomos Desktop roadmap.

Native AI desktop apps (Claude Desktop, ChatGPT Desktop, Copilot, Perplexity Desktop)

Not supported in the browser extension. A browser extension fundamentally cannot see traffic from a native app. This is the single biggest gap the prototype has, and the primary motivation for Sonomos Desktop, which integrates with these clients directly.

IDE AI assistants (Cursor, Windsurf, Copilot Chat, JetBrains AI, Continue)

Not supported in the browser extension. IDE assistants typically route through native processes outside the browser. Coming in Sonomos Desktop.

Claude Code

The browser extension can’t reach Claude Code either — but Canary can. Canary is Sonomos’s open-source PII monitor for Claude Code that installs as a plugin with two slash commands. It’s available today and runs entirely locally.

Local LLMs (Ollama, LM Studio, llama.cpp, vLLM)

Not supported in the browser extension. Even though “local” suggests safety, local LLMs still keep prompts in chat history, logs, and disk cache — Sonomos Desktop will let you scrub PII before it lands there.

Best practices

• Always enable Cloak on supported platforms — it’s the most seamless protection.
• Don’t disable Send Guard — it’s your last line of defense on uncovered sites.
• Check the risk widget before submitting — a red indicator means high-severity PII is present. If you’re surprised by red, look at what Dagger detected before you continue.
• Prefer placeholders over real data even in your own prompts, especially when the AI doesn’t need the real value to be useful.
• Use compliance reports (Professional / Teams) to document your PII handling for audits.

What’s next?

Cloak reference Deep dive on how interception works and what its limits are.

Sonomos Desktop coverage See which native apps, IDEs, and local LLMs are coming.