Privacy Policy

Henji (the "Service") is operated by Appelier, a sole proprietorship ("we", "us", or "Operator"). This Privacy Policy describes how we handle information about you (the "Customer" or "you") when you use the Service. It is incorporated into our Terms of Service.

Last updated: July 20, 2026

1. Scope

This Policy applies to the Henji desktop application (macOS), the iOS app and the Henji Keyboard (keyboard extension), the landing page (https://henji.ai), the backend API, and related services.

2. Information We Collect

2.1 Collected Automatically

Device identifiers: an auto-generated Device ID and a hash of a device-specific value (macOS: a SHA-256 hash of the hardware UUID; iOS: a hash of an Apple App Attest key ID). We never store the raw device identifier.
Usage statistics: number of launches, generations, agent names, input/output token counts.
App / OS info: app version, macOS / iOS version, locale.
Error logs: crash information, API error codes.

2.2 Provided by You

Inputs to AI Features (macOS): screenshot images (OCR-processed on your device; see §2.4), text instructions, voice instructions (transcribed on your device and handled as text; the audio itself is never transmitted or stored — see §2.4), agent configurations.
Inputs to AI Features (iOS): text you copied (we read the clipboard only when you explicitly tap to capture it), text you selected in an input field (only when you run a proofread), text instructions, voice instructions (transcribed inside the iOS app and handled as text), and agent configurations.
Outputs: AI-generated reply candidates — displayed locally on your device. We do not store the body of Outputs on our servers.
Payment information (macOS): entered directly on Stripe Checkout. Card numbers and similar payment data never pass through our servers.
Payment information (iOS): purchases use App Store In-App Purchase (StoreKit). Payment details are handled by Apple and are never received by us. We receive only the transaction identifier and its status, which we use to confirm your subscription.
Account information (optional): if you sign in with Apple or Google, your user identifier and email address. Signing in is optional — you can skip it and still use every feature.
Contact info on support / limit-increase requests: email address and a description of your usage.

2.3 What We Do NOT Collect

The following are never collected or transmitted (including to our analytics backend, Google Analytics):

Your keystrokes. The Henji Keyboard does not collect or transmit what you type. The keyboard sends text to the network only when you explicitly tap "generate a reply" or "proofread" — and only the text that action targets.
Clipboard contents read in the background (we read the clipboard only when you explicitly capture it)
Screenshot images (macOS; see §4)
Recorded audio itself (on both macOS and iOS, transcription happens entirely on your device; audio is never sent to our servers or to the AI model provider, and is never written to a file)
The body of generated reply text
The body of user instruction text
The raw device identifier (we keep only a hash)

2.4 macOS Permissions Requested by the Service

The Service requests the following macOS permissions when you choose to use the relevant features. You may revoke any of them at any time from System Settings (the corresponding feature will then be unavailable).

Permission	Purpose	What we access	Where it goes
Screen Recording	Capturing screenshots	The captured screenshot image, converted to text on-device via Vision OCR	Only the OCR text is sent to the AI model provider (§4). The image itself never leaves your device.
Accessibility	Detecting text selection, generating replies, and recording the source app	The text you have selected and its source app name (e.g., `Slack`, `Mail`)	The selected text and its source app name are sent to the AI model provider (§4) only when you click the icon to generate a reply.
Microphone & Speech Recognition	Turning what you say into a reply instruction (voice input)	Microphone input, only while you have started a recording	Never transmitted. Transcription runs entirely on your device using macOS speech recognition, so the audio never leaves it. The resulting text instruction is sent to the AI model provider (§4) only if you go on to generate a reply.

We collect or transmit data tied to a permission only when that permission is granted. Selected text is never sent off-device unless you actively trigger a reply generation.

2.5 iOS Permissions and "Full Access"

The iOS app uses the following permissions. You can revoke any of them at any time in the Settings app.

Permission	Purpose	What we access	Where it goes
Keyboard Full Access	Network access for AI reply generation / proofreading, clipboard capture, shared settings	Text you explicitly captured, or text you selected in an input field	AI model providers (§4) — only when you run a generation or a proofread
Paste from Other Apps (optional)	Removes the repeated paste confirmation when capturing a copied message	Clipboard text (only on explicit capture)	Same as above
Microphone / Speech (optional)	Voice custom instructions	Audio is transcribed inside the iOS app	Audio itself is never transmitted. Only the transcribed instruction goes to §4.

Important notes about Full Access:

The Henji Keyboard works as a quick-reply keyboard without Full Access. Full Access is required only for AI reply generation, proofreading, and voice.
Even with Full Access granted, we never collect or transmit your keystrokes.
The keyboard extension transmits only the text targeted by an explicit "generate a reply" / "proofread" tap.
The keyboard extension never presents in-app purchases (purchases happen only inside the iOS app).

3. Purposes of Use

We use the information above for:

Delivering the Service (executing generation requests, displaying replies)
Billing (Stripe-based payments and subscription management)
Enforcing monthly usage limits
Quality and reliability improvement (aggregated, anonymized statistics)
Responding to support inquiries and limit-increase requests
Detecting and addressing violations of the Terms
Complying with legal obligations

4. How AI Features Handle Inputs and Outputs

What is transmitted (macOS): Screenshots are OCR-processed locally on your device via macOS Vision. When you use AI Features, only the extracted text and your instructions are transmitted over HTTPS to our AI model provider (Google LLC) which returns Outputs. The screenshot image itself is never sent to our servers or to the AI model provider.
What is transmitted (iOS): The message you explicitly captured (from the clipboard) or the text you selected in an input field, together with your custom instruction and your style rules, is transmitted over HTTPS through our backend (api.henji.ai) to our AI model provider (Google LLC (Gemini)). Transmission happens only at the moment you run "generate a reply" or "proofread." Keystrokes are never transmitted. We obtain your explicit consent for this transmission beforehand, on a consent screen shown during onboarding that names what is sent and who receives it. If you decline, AI features stay disabled and you can still use Henji as a quick-reply keyboard.
Not stored by us: We do not store the body of Inputs or Outputs in our backend DB (Cloudflare D1). What we store is metadata: usage counts, token counts, agent IDs, etc.
Provider-side retention: We have enabled the no-logging / zero-data-retention options offered by our AI provider. Google does not use Inputs or Outputs to train its models and does not retain them beyond brief operational processing windows, in accordance with its published policies.
Encryption in transit: All transit is encrypted with TLS 1.2 or higher.
Local history: Where the Service offers a local generation history, those records remain on your device and are not transmitted to our servers.

5. No Training on Your Data

We do not use your Inputs or Outputs to train, fine-tune, or evaluate our own or any third party's AI models. We have also enabled the no-training opt-outs available from our AI Sub-Processors.

6. Sub-Processors

We rely on the following Sub-Processors to deliver the Service.

Category	Recipient	Data shared	Purpose	Primary locations
AI model	Google LLC (Gemini)	Extracted text (macOS: OCR result / iOS: captured message or selected text) and instructions (no image)	Reply generation	United States
Payments	Stripe, Inc. (macOS)	Card info, email, billing history	Payments and subscription management	US, EU, Japan
Payments	Apple Inc. (iOS)	App Store in-app purchase transaction data (payment details never reach us)	In-app purchase and subscription state	US, Japan
Cloud	Cloudflare, Inc.	Device ID, usage stats, subscription state	Backend / API hosting and database	US, EU, Japan regions
Analytics	Google LLC	Device ID, usage stats, app / OS version	Anonymous analytics	United States

Sub-Processors may be added or replaced from time to time; material changes will be reflected in the updated date on this page.

7. Data Storage and Retention

Data	Location	Retention
License info	Keychain (local to your device; macOS / iOS)	Until you delete it
Reply history, personas, quick replies, memories (iOS)	Shared on-device database (App Group) — local only	Until you delete them or delete the app
Device ID, subscription state	Cloudflare D1 (US, EU, Japan)	12 months after cancellation, or upon deletion request
Monthly usage counts	Cloudflare D1	25 months (to support year-over-year comparison)
Payment info	Stripe (macOS) / Apple (iOS)	Per their respective policies
Analytics	Google Analytics (US)	14 months (default retention)
AI Inputs / Outputs	(not stored by us)	—

8. International Data Transfers

Your data may be processed by Sub-Processors located in the United States, the EU, Japan, and elsewhere. We rely on appropriate technical and organizational safeguards (TLS encryption, access controls, no-logging settings, etc.) to keep your data secure across regions.

9. Your Rights

You may request to:

Access the information we hold about you
Correct inaccurate information
Delete your information
Restrict specific processing
Port your data in a machine-readable format

See §11 for how to make these requests.

10. Security

We apply reasonable technical and organizational measures to protect your information against unauthorized access, alteration, disclosure, and loss, including:

TLS encryption in transit
Local encryption of sensitive data using the Keychain (macOS: Electron safeStorage; iOS: shared Keychain)
No AI provider API keys on your device — AI calls always go through our backend
Access controls and least-privilege principles
Avoiding unnecessary data retention (we do not persist the body of Inputs or Outputs)

11. Deletion Requests and Contact

For deletion requests, questions about this Policy, or to exercise your rights:

Contact: support@henji.ai
Please include: the last 8 characters of your Device ID, shown in the Settings screen.
Response: we respond within 14 days. Items we can delete immediately (e.g., Google Analytics events) are handled within 24 hours.

12. Cookies

Our landing page (https://henji.ai) may use cookies or similar technologies to remember your language preference (hl=ja|en) and to collect anonymized usage data via Google Analytics.

13. Analytics Settings

In-app analytics is always enabled. The data sent is limited to what is described in §2.1 and never includes the body of Inputs or Outputs.

14. Children's Privacy

The Service is not directed to children under 13 (or under 16 in jurisdictions that apply that threshold). We do not knowingly collect information from these individuals.

15. Updates

We may update this Policy as laws change, as the Service evolves, or as we update Sub-Processors. Material changes will be announced via in-app notice or email.

16. Revision History

Date	Change
2026-05-10	Initial publication
2026-05-13	Restructured to mirror Notion AI's Privacy Policy: AI Inputs/Outputs handling, explicit no-training statement, Sub-Processor list, retention schedule, international transfers, and user rights
2026-05-28	Added §2.4 "macOS Permissions Requested by the Service" (Screen Recording / Accessibility — usage and transmission scope)
2026-05-31	Updated §4 and §6 to match implementation: clarified that only OCR-extracted text (not the image) is sent to AI model providers, and corrected the providers to Anthropic / Google
2026-06-07	Aligned with implementation: removed "format preset names" from §2.1 usage statistics (the format-preset feature was discontinued), and revised the §2.4 Accessibility purpose from "active-app detection" to recording the selection's source app as provenance / context
2026-07-14	Brought iOS (the app and the Henji Keyboard) into scope. Added §2.5 "iOS Permissions and Full Access" (stating that keystrokes are never collected and that transmission happens only on your explicit action), and updated §2.2, §2.3, §4, §6, §7, and §10 to match the iOS implementation (clipboard / selected-text capture, App Attest, App Group local storage, Apple in-app purchases)
2026-07-28	Added macOS voice input: recorded a Microphone & Speech Recognition row in §2.4, noted in §2.2 that voice instructions are transcribed on-device, and stated in §2.3 that recorded audio itself is never collected or transmitted
2026-07-20	Corrected the AI model provider in §4 and §6 to Google LLC (Gemini) only (reply generation on both macOS and iOS goes through our backend and uses Gemini only; Anthropic is not used, so it was removed from the sub-processor list and the body text)

Contact: support@henji.ai