Documentation
Everything you need to install, configure and use Gate Kali — the browser extension and the assistant.
Gate Kali ExtensionBrowser extension & sidecar — install, connect, protect your prompts
1. Installation
Option A — Chrome Web Store
- Open the Gate Kali listing in the Chrome Web Store.
- Click Add to Chrome.
8a7d5921fca6d3c0e729e638d5d838d1c4f54afd671e8fac4249d825354399a7Use this when configuring managed Chrome policy or verifying the extension.
Option B — Sideload (developer mode)
- Download gate-kali-extension.zip and extract it.
- Open
chrome://extensionsin Chrome. - Enable Developer mode (top right).
- Click Load unpacked and select the extracted folder.
2. Connect the sidecar
Gate Kali needs a sidecar endpoint that performs the sanitization. You can use the hosted endpoint or your own.
- Click the Gate Kali toolbar icon to open the popup.
- In Sidecar endpoint, set the API base — e.g.
https://api.gatekali.com. - Paste your client token in the token field.
- Click Save endpoint. Chrome will ask for host permission — approve it.
- Click Test next to the status line. You should see Sidecar connected · api.gatekali.com.
8a7d5921fca6d3c0e729e638d5d838d1c4f54afd671e8fac4249d825354399a7You may need this when approving host permission for
https://api.gatekali.com or setting managed Chrome policy.
3. How sanitization works
When you interact with a supported AI site, Gate Kali intercepts content before it leaves the page:
- Send: The prompt is sent to the sidecar, which detects PII and replaces values with tokens like
[EMAIL_005]or[US_SSN_001]. Only the sanitized text reaches the AI. - Paste: Content pasted into the composer is sanitized the same way.
- File upload: Attached files are scanned and sensitive values are tokenized.
- Warnings: Prompt-injection patterns are flagged as warnings — your prompt is still sent (it's yours), but you're told what was detected.
4. Feature guide
| Popup control | What it does |
|---|---|
| Protection enabled | Master on/off switch for all sanitization. |
| Zero-transit (on-device) | Runs the bundled WASM engine locally — no network call per prompt. |
| Recent activity | Local log of sanitized sends/pastes/uploads with token counts. Clear wipes it. |
| Protected sites | Built-in AI sites plus any custom domain you add via Protect another site. |
| Redact custom terms | Your own sensitive words/phrases (codenames, domains) are tokenized like PII. |
| Sidecar endpoint | API base URL + client token. Test re-checks reachability. |
| Toolbar badge | Green dot = connected; red = unreachable; OFFLINE = no sidecar configured. |
5. Supported sites
Gate Kali protects prompts on these platforms out of the box:
- ChatGPT (chat.openai.com, chatgpt.com)
- Claude (claude.ai)
- Gemini & AI Studio (gemini.google.com, aistudio.google.com)
- Perplexity (perplexity.ai)
- Grok & X (grok.com, x.com)
- Microsoft Copilot & Bing (copilot.microsoft.com, bing.com)
- DeepSeek (deepseek.com)
- Meta AI (meta.ai)
- Mistral Le Chat (chat.mistral.ai)
- Hugging Face (huggingface.co)
- Poe (poe.com)
Anything else? Type the domain into Protect another site in the popup.
6. Zero-transit (on-device) mode
Enable Zero-transit in the popup to sanitize entirely in the browser using the bundled WASM engine:
- No network round trip per prompt — works even if the sidecar is unreachable.
- Same detector set as the sidecar for common PII types.
- Advanced features (admin-managed custom detectors, cross-session vault) still need the sidecar.
7. Troubleshooting
| Symptom | Fix |
|---|---|
| "Sidecar not reachable" / red badge | Check the endpoint URL, click Test, and confirm curl https://your-sidecar/health returns {"status":"ok"}. |
| "Sidecar required" banner | Set an endpoint + token, or enable Zero-transit mode. |
| Permission error on save | Approve the host-permission prompt — it needs a user gesture, so click Save endpoint again. |
| "Update required" banner | Extension and sidecar protocol versions differ — update whichever is older. |
| Nothing is sanitized | Check Protection enabled is on and the site is in Protected sites. |
| A prompt was blocked | Injection heuristics fire warnings, not blocks, on your own prompts. A hard block means the sidecar policy denied it — check your sidecar admin dashboard. |
8. FAQ
Does Gate Kali see my prompts?
Only the sidecar endpoint you configure sees them — and only for the instant it takes to sanitize. Gate Kali collects no analytics.
Can I use it without a sidecar?
Yes — enable Zero-transit mode for fully on-device sanitization.
How do I redact a codename or internal domain?
Add it under Redact custom terms in the popup.
Can I generate my own token?
Yes — tokens are just secrets your sidecar accepts. Ask your sidecar admin, or if you run the sidecar, mint one in the admin dashboard.
Gate Kali Assistant BetaSelf-hosted, multi-tenant AI principal assistant — access, integrations and governance
Gate Kali Assistant reads your inbox, calendar and documents — then briefs you, prepares meetings, triages correspondence and drafts replies. Every consequential action waits for your approval. Full capability overview: assistant product page.
Getting access
- Go to mygate.gatekali.com and sign up with your email — public signup provisions an isolated tenant workspace with an admin account.
- Verify your email, then sign in to your workspace dashboard.
Connecting your world
- Google: connect Gmail, Google Calendar and Google Drive via OAuth consent — you approve the exact scopes (see the privacy policy for how Google data is handled).
- Other email: generic IMAP/SMTP and Apple iCloud are supported.
- Chat channels: link WhatsApp, Telegram, Slack, Teams or Discord via link codes or webhooks.
- Documents: upload files to the knowledge base for retrieval in chats and briefings.
Working with the assistant
- Chat & lenses: the assistant auto-detects a context lens (Personal, Diplomatic, Government, Security, Media, Commercial, Social Protocol) to shape tone, priorities and tool access.
- Briefings: ask for a briefing, or configure scheduled daily briefs and heartbeat checks for inbox, calendar and follow-ups.
- Approvals: reads run freely; sends, deletes and shares pause for your explicit approve/reject. Every decision is audit-logged.
- Meeting bots: the assistant can auto-join meetings to record, transcribe and summarize.
- Real-time: replies, reminders and notifications push live over WebSocket.
Administration
- Tenant branding (name, logo, accent colors), feature flags and privacy mode are managed from workspace settings.
- Operators self-hosting the platform: see Operators & Self-Hosting and the platform README (Docker Compose,
docker-compose.prod.ymlwith nginx + TLS).
Support
- Issues & bugs: github.com/Evansone001/gate-kali/issues
- Email: evans@nacextech.com
- Privacy: privacy policy