ArcSolve Engine Privacy Policy

This is the text of a specific version. It is the version in effect.

English

ARC-SOLVE (“we”) explains in this policy what information the Mac app ArcSolve Engine (“Engine”) handles. Engine is a menu bar app that runs local AI models on your Mac. AI agents on the same Mac give Engine transcription and speech synthesis work through MCP or local HTTP.

This policy applies to Engine and to inquiries you send us directly about Engine. Visits to our websites that carry the Engine help and this policy are described in Visits to our websites. See Websites and external links for pages opened by the app menu. If the English text differs from the Korean text, the Korean text prevails.

Information we handle

Engine’s inference feature processes agents’ audio and text, and the results, on your Mac. It does not send them to us.
Engine has no sign-up, sign-in, advertising, built-in usage analytics, or tracking. It contains no third-party SDK for those purposes. Engine does not use cookies.
Engine uses connections from agents on the same Mac (127.0.0.1) and Apple’s model delivery. When Engine is installed or started, macOS may check the list of available models with Apple. macOS downloads model files only for the models you choose. Apple model delivery requests do not contain your input or results. See Model downloads.
For company inquiries and websites outside the apps, we process information you send us directly and web access information that passes through Cloudflare. See Information we handle when you contact us and Visits to our websites. We receive and use crash reports and app usage statistics provided by Apple. See Crash reports and app usage statistics provided by Apple below.

Engine processes the information below on your Mac for its features.

Information	Purpose	Where it is stored
Input from agents: audio files to transcribe, text to speak, and voice and language values	Transcription and speech synthesis you request	Memory of the running Engine, Engine’s work folder on your Mac
Results: transcribed text, generated audio files	Returning results to the requesting agent	Memory of the running Engine, Engine’s work folder on your Mac
Connection information: connection credentials (tokens), caller IDs, model management permission, port, folder locations	Agent authentication and running Engine	Engine’s settings file on your Mac
Installed models: model files, install records	Running models	Engine’s model folder on your Mac

Audio and text can contain personal information that you or an agent put in them. Engine uses this content only for the requested processing and for returning the result.
On its first launch, Engine creates the settings file and a random connection credential in the app’s storage on your Mac. When you choose “연결 정보 복사” (copy connection information) in the menu, Engine puts the address and the credential on this Mac’s clipboard. This item does not go to your other devices through Universal Clipboard.
When an agent asks to download a model, Engine downloads nothing yet and shows “에이전트가 요청함” (requested by an agent) on that model’s row in the menu. macOS asks you to allow notifications when the first request arrives. If you allowed notifications, Engine sends one macOS notification for each requested model. It contains only fixed guidance text, the model name and the download size. Engine removes it when the request leaves approval waiting after download approval, decline, or cancellation. The download starts when you choose “받기” (download) on that row. If you choose “거절” (decline), nothing is downloaded.
Engine does not write input, results or connection credentials to logs. For troubleshooting, it writes only fixed error codes and the identifier of a model that failed to download to your Mac’s system log. Engine itself has no feature that uploads this log to us.

Agents that receive results

Engine returns a result to the agent that requested the job with that job’s connection credential.
On its first launch, Engine creates one connection credential (caller ID local-agent). Engine does not distinguish agents that share the connection details. If they know a job identifier, they can read each other’s jobs and results and cancel unfinished jobs. This credential also has model management permission. Agents with it can request model downloads and remove installed models that are not in use without asking you. A download starts only when you choose “받기” (download) in the menu. Give the connection details only to agents you trust.
How an agent and its AI service handle a result follows that service’s policy. You choose which agents to connect.

Model downloads

Engine from the App Store downloads model files from Apple’s servers. macOS may check the model list at installation or startup. When you choose “받기” (download) in the menu, macOS downloads the model pack and hands it to Engine. A request from an agent alone does not start a download.
The Engine app has no permission to open connections to outside servers itself. Model download requests do not contain your input or results.
These downloads and the information that the App Store and macOS process follow the Apple Privacy Policy.
Engine uses a downloaded model file only after it matches a fixed hash.

Retention and deletion

Information	Retention	How it is deleted
Input and results	Result files: until Engine stops or completed jobs exceed a fixed limit. Input files: until Engine stops	When you quit the app or Engine restarts by itself, Engine deletes the work folder (input and result files) and the job records. While it runs, Engine deletes the result files of the oldest jobs when completed jobs exceed a fixed limit. Engine does not delete input files while it runs
Connection information	Until you delete the settings file	You delete the settings file in the app’s storage. Engine creates a new connection credential on the next launch
Model files and install records	Until the model is removed	Engine deletes them when you choose “지우기…” (remove) for that model in the menu, or when an agent with model management permission requests removal

If Engine cannot end normally, for example after a force quit or power loss, or if cleanup does not finish within 10 seconds of quitting, files can remain in the work folder. Engine tries to clean them up at the next start.
At the next start, Engine cleans up remaining job files and partly downloaded model packs. If cleanup fails, files can remain. You can delete job files remaining in Engine’s app storage yourself. macOS handles storage and deletion of model packs managed by Apple.

Websites and external links

The menu items “개인정보 처리방침” (privacy policy) and “모델 카드 열기” (open model card) open the web page in your Mac’s default web browser. The Engine app does not send the request itself.
“모델 카드 열기” (open model card) opens the model repository page under mlx-community on Hugging Face. This community repository provides an MLX conversion. What happens next follows the policies of your browser and Hugging Face.
Website visits are separate from what the Engine app processes. Engine does not send audio, text or results to any website.

Crash reports and app usage statistics provided by Apple

These statements apply to App Store builds. If you enable analytics sharing and sharing with app developers, Apple may collect this app’s crash reports and usage statistics. Apple may provide these materials to us. We receive and use these materials from Apple.

Crash reports may include app and operating system versions, device information, and call stacks of the code that crashed. App usage statistics may include session counts, time in the app, and active device counts. We use these materials to analyze crashes, fix errors, understand app usage, and improve app quality.

We view these materials only in App Store Connect and Xcode. We do not download them for separate storage. Xcode may automatically download local copies for viewing. Retention and deletion of these materials follow Apple’s policies. We do not link these materials to you or use them for tracking.

Collection, sharing, retention, and privacy protections within Apple’s services follow Apple’s Privacy Policy and Apple’s App Analytics information.

To refuse future sharing, open Settings on iPhone or iPad, or System Settings on Mac. Under Privacy & Security → Analytics & Improvements, turn off sharing with app developers. See iPhone and iPad analytics information and Mac analytics sharing information for your choices. Stopping sharing does not mean that materials already provided are deleted.

Device information and deletion

We cannot see the input, results, connection information or models that Engine handles on your Mac. You check and delete them yourself as shown in Retention and deletion.

Safeguards

If data categories or purposes change, we will update this policy and the App Store privacy details before the change.
Engine accepts only connections from the same Mac (127.0.0.1) and rejects requests without a registered connection credential.
Engine separates jobs and files by the caller ID of the connection credential. Jobs and files of another caller ID cannot be read. Engine creates only one credential on first launch, so agents that share it are not separated from each other. See Agents that receive results.
Engine creates the settings file so that only its owner can read and write it, and accepts a settings file only when no one other than its owner has any access to it. Only the owner can open the work folder.
Engine runs inside the macOS App Sandbox.
Engine uses a downloaded model file only after it matches a fixed hash.

Children

Engine has no accounts and does not ask your age.

Product inquiries

The Engine app has no feature for sending inquiries. When you contact us yourself by email or phone, we handle the information in Information we handle when you contact us.

Do not include original audio or documents, connection credentials, passwords, or unnecessary personal records in your inquiry.

Adding [ArcSolve Engine] to the start of the email subject helps us respond faster.

Operator and privacy contact

The operator is 아크솔브(ARC-SOLVE). Our privacy officer is 조경민 (Kyungmin Cho), the representative. Send privacy questions and rights requests to the contact below.

Email: support@arcsolve.ai
Phone: 050-6508-1228 (Korea)
Business name: 아크솔브(ARC-SOLVE) · ArcSolve
Representative: 조경민(KYUNGMIN CHO)
Business registration number: 843-54-01035
Address: 서울특별시 성북구 장위로19길 25, 1동 302호(장위동, 장위 에스하임1), Seoul, Republic of Korea

Information we handle when you contact us

We handle the information below when you contact us directly by email or phone.

Case	Information	Purpose	Retention
Email inquiries	Email address, sender name, email header information (such as the time sent), and the message and attachments	Reviewing and answering your inquiry and following up on it	One year after we finish handling the inquiry
Phone inquiries	Phone number and what you tell us	Reviewing and answering your inquiry	One year after we finish handling the inquiry

We process inquiry email through the processors below. We do not provide inquiry information for another company’s independent use, except where the law requires it.

Deleting inquiry information

Procedure: When the retention period ends, we delete inquiry emails and records without delay. If a law requires longer retention, we keep them separately from other records for that period and then delete them.
Method: We delete electronic records, and delete inquiry emails from the mailbox and its trash as well.
Our deletion of email is separate from completion of deletion within Google’s systems. Google’s deletion process and time limits follow the customer data deletion provisions of the Google Data Processing Addendum.

Processors

We use Google Workspace to receive and store email for support@arcsolve.ai. We use Cloudflare to serve our websites and process web requests. These processors handle company inquiry email and websites. This processing does not send documents, model inputs, or tool results from the apps to Google or Cloudflare. Content you put in an inquiry is processed as inquiry email.

Service provider	Task	Information processed
Google (Google Workspace)	Receiving and storing support email	Email address, sender name, email header information, and the message and attachments
Cloudflare	Serving static websites and processing web requests	IP address and the request information sent by the browser

Google’s public contracting entity table identifies Google Asia Pacific Pte. Ltd. for Google Workspace contracts with a Korean billing address. The Cloud Data Protection Team is a channel for our Workspace administrator to contact Google about customer data. Send your privacy inquiries and rights requests to our company contact above. We contact Google through our administrator when needed.

Google’s role in processing customer data is described in the Google Data Processing Addendum. Cloudflare’s processing role is described in the Cloudflare Data Processing Addendum.

Processing outside Korea and your choices

Google and Cloudflare may use facilities and subprocessors in multiple countries under their respective data processing addenda. Inquiry email and web access information may be processed outside Korea while these services are provided.

The table above lists the inquiry email information processed through Google Workspace. It is sent over the network when email is received. We retain it for one year after we finish handling the inquiry.
If you do not want to use the Google Workspace email route, contact us by phone instead of email. In that case we do not answer by email.
Cloudflare processes the IP address and browser request information sent over the network when you open our web pages. Web access information is used to serve pages and for security processing.
Google’s processing locations and subprocessors are described in the Google Workspace subprocessors information. Cloudflare’s subprocessors are described in the Cloudflare subprocessors information.

Visits to our websites

The websites that provide our company and product introductions, help, and legal documents are static pages. The sites themselves use no cookies, advertising, analytics, or tracking scripts. When you open a page, your IP address and the request information your browser sends pass through Cloudflare’s servers. Our site code has no feature that stores this access information separately.

Your rights and how to use them

You can ask to access, correct, or delete the personal information we handle about you (your inquiry records), or to stop processing it. Send your request to the contact above. We will explain how we confirm your identity and then handle the request.
A legal representative or a person you authorize can also make the request.

Safeguards for our websites

Our websites are served over HTTPS. Their Content Security Policy restricts scripts from outside origins, additional network connections from pages, and form submissions.

Inquiries from children

Children under 14 should contact us together with a legal guardian.

Privacy violation assistance

You can also report or ask about a privacy violation to these Korean public bodies.

Personal Information Infringement Report Center: privacy.kisa.or.kr, 118 (in Korea)
Personal Information Dispute Mediation Committee: www.kopico.go.kr, 1833-6972 (in Korea)

Changes to this policy

When this policy changes, we publish a new version and its effective date. If earlier versions have been published, they are available in the change history.