Message blocked
Confidential Agent · Data protection
API keyPasswordInternal URL
Risk score
Review redacted text or cancel.
Anonymized preview — review before sending
[REDACTED_INTERNAL_URL] … [REDACTED_API_KEY] … [REDACTED_PASSWORD]
Browser-native · AI · Social · Youth
Nothing toxic, secret,or careless ships.
Intercept prompts, posts, and DMs before they leave the tab — for teams, brands, and young people online.
Explainable · Auditable · Yours
AI
Social
Youth
Same keyboard. Every surface.
Social
Social networks — first-class
Feeds, stories, replies & DMs: same exfiltration and reputation risk as corporate AI.
XLinkedInInstagramTikTokDiscordReddit
AI
Generative AI
Secrets in pasted code & prompts — blocked at submit.
POST /v1/evaluate → block · mask · OK
Youth
Minors & digital dignity
Vulgarity, harassment, unsafe content — interrupted with calm, age-aware prompts.
VulgarityHarassmentAge-aware
Four jobs. One layer.
Social surfaces
Public threads & private messages — one policy.
AI copilots
DLP for everything developers paste into chat.
Tone & vulgarity
Harsh or explicit language flagged before send.
Audit trail
Incidents your security team can prove.
How it works
Three steps. No lecture.
01
Install
Extension + your API URL.
02
Block at send
We score text in the browser.
03
Log & teach
Allow · mask · warn · block.
Four ways words hurt
Leaks
Keys & code pasted into AI or social.
Toxic threads
Harassment that scales in public DMs.
Vulgarity & harm
Especially around minors — one send, lasting damage.
Compliance
You need logs, not screenshots.
The fix
Stop. Explain. Offer a path.
- Plain-language reason on every block.
- Redact, rephrase, or pause — you choose.
- Dashboard for security & families.
Live UI
What people see
Data leak vs. tone — same calm chrome.
Language improvement suggested
Confidential Agent · Tone moderation
Harsh or harmful tone
Pick a softer option or edit.
How can we address this respectfully and de-escalate?
What outcome do we need — and what’s the kindest path there?
Could we reframe this to protect collaboration?
Roles
Built for the people on the hook
Same calm overlay — whether the risk is a leaked secret or a harmful DM.
Security & IT
Security
Catch accidental exfiltration in AI and internal tools — incidents your board can actually read.
Secrets & DLPAudit trail
HR & education
HR & schools
Shield minors and staff before vulgarity, harassment, or escalation leaves the tab — without surveillance theater.
Minors onlineVulgarity & tone
Brand & community
Community
One policy for public threads, stories, and DMs — brand-safe replies at scale.
Social surfacesModeration
Developers
Source code isn’t “just text”
Paste into AI or forums = data on someone else’s infra.
Repo snippets & .env
Blocked or masked before HTTP leaves the browser.
Shadow AI
Same rules on every “quick ask”.
Audit
What, why, when — exportable.
Serious infrastructure underneath
Operational truth in MongoDB, semantic depth with Qdrant and OpenAI embeddings — designed to extend, not lock you in.
Read the stack overview →Install the extension
Choose your browser. Use the developer build until store listings are live.
Chrome
SoonUse developer unpacked build until listed.
Store hub (replace when listed) →Edge
SoonEdge accepts unpacked MV3 similarly to Chrome during development.
Store hub (replace when listed) →Firefox
SoonFirefox MV3 packaging may differ; track in repo issues.
Store hub (replace when listed) →Safari
SoonSafari Web Extensions workflow is separate from Chromium.
Store hub (replace when listed) →Developer (unpacked)
Load apps/browser-extension from chrome://extensions in developer mode.
↓Connect the extension to your API
In extension options, set the API base URL (e.g. http://localhost:8080). Add an API key from the dashboard when enforcement is enabled.
- Install the extension for your browser.
- Run security-api locally or use your deployed URL.
- Paste the API URL (and API key if required) in extension options.
Give every teammate a safety net they trust
Install Confidential Agent, connect your API, and replace anxious moments with teachable, logged decisions.