myllm | Opticell Ltd.

# MyLLM — Privacy Policy

**Last updated:** March 17, 2026

**Effective date:** March 17, 2026

## Summary

MyLLM does not collect, transmit, store, or share any personal data with us or any third party. Everything stays on your device and your local network.

---

## 1. Data Collection

**We collect no data whatsoever.** MyLLM has no analytics, no telemetry, no crash reporting, no user accounts, and no server-side infrastructure operated by us.

Specifically, MyLLM does **not** collect:

- Personal information (name, email, phone number, address)

- Device identifiers (IDFA, IDFV, serial numbers)

- Usage analytics or behavioural data

- Crash logs or diagnostics sent to us

- Location data

- Browsing or search history

- Financial or payment information

## 2. How the App Works

MyLLM is a local AI assistant that processes all data on your device and your private network. Understanding the architecture is key to understanding the privacy model.

### 2.1 Server Mode (Ollama, LM Studio, Custom)

In server mode, MyLLM connects to a Large Language Model (LLM) server that **you** operate on **your** local network — typically a computer in your home or office running software such as Ollama or LM Studio.

- **Your messages** are sent from the MyLLM app on your iPhone/iPad to your own server over your local network (e.g., 192.168.x.x).

- **AI responses** are generated by the LLM running on your server and returned to the app over the same local connection.

- **No data leaves your network.** The communication path is: your iPhone → your WiFi router → your computer. At no point does any data pass through our servers, cloud services, or the public internet (unless your server is intentionally configured to be internet-accessible, which is your responsibility).

- **Tool results** (web searches, file contents, contacts, calendar events, photos) are retrieved by the app on-device and sent to your local server for processing. They are never sent to us.

### 2.2 On-Device Offline Mode

In offline mode, MyLLM runs a language model directly on your iPhone or iPad using the llama.cpp inference engine with Metal GPU acceleration.

- **Model files** are downloaded from Hugging Face (huggingface.co) to your device's local storage. This is the only external network request the app makes in offline mode, and only when you explicitly tap "Download." Hugging Face's own privacy policy governs that download.

- **After download, no internet connection is needed.** All inference runs entirely on your device's processor and GPU.

- **Your messages and responses never leave the device.** There is no server involved.

### 2.3 Large Language Models (LLMs)

MyLLM does not provide or operate any AI model. The app is a client that connects to models you choose to run:

- **In server mode**, the model runs on hardware you own and control. The model's behaviour, capabilities, and any data it processes are governed by your server configuration and the model's license.

- **In offline mode**, small open-source models (SmolLM, Qwen 2.5, TinyLlama) run directly on your iPhone. These models are open-weight and do not phone home or transmit data.

- **We have no access** to the prompts you send, the responses you receive, or any data processed by the model — whether on your server or on your device.

### 2.4 What the AI Can Access (With Your Permission)

When you grant iOS permissions, the AI agent can access:

- **Contacts** — to search and look up contact information on your behalf

- **Calendar & Reminders** — to view and create events and reminders

- **Photos** — to browse and save images

- **Microphone** — for voice input (speech-to-text processed by iOS on-device)

- **Camera** — to capture photos to share with the AI

- **Files** — to read, write, and manage documents in the app's sandbox

All of this data is processed locally. Contact names, calendar events, photo metadata, and file contents are sent only to your own local LLM server (or processed on-device in offline mode) — never to us or any external service.

### 2.5 Web Search and URL Fetching

When you ask MyLLM to search the web or fetch a URL, the app makes standard HTTPS requests from your device to the public internet (e.g., search engines, news sites, or URLs you specify). These requests come from your device's IP address and are subject to the privacy policies of those websites. We do not proxy, log, or intercept these requests.

### 2.6 Server Terminal

The Server Terminal feature allows you to manage your Ollama server (pull, delete, inspect models) remotely from your iPhone. All commands are sent directly from your device to your server over your local network. We have no visibility into these operations.

## 3. Data Storage

All app data is stored locally on your device:

- **Conversations** — saved in the app's sandboxed documents directory

- **Memories** — user-taught facts stored in local JSON files

- **Workflows** — automation definitions stored locally

- **Downloaded models** — GGUF files stored in the app's documents directory

- **Settings** — stored in UserDefaults and Keychain (for API keys)

No data is synced to iCloud, backed up to our servers, or transmitted externally. If you delete the app, all data is permanently removed from your device.

## 4. Third-Party Services

MyLLM does **not** integrate with any third-party analytics, advertising, crash reporting, or tracking services. Specifically:

- No Google Analytics, Firebase, or Crashlytics

- No Facebook SDK or Meta tracking

- No ad networks or attribution services

- No Mixpanel, Amplitude, or similar analytics

- No Sentry, Bugsnag, or similar error reporting

The only external network requests the app makes are:

|---|---|---|---|

## 5. Children's Privacy

MyLLM does not knowingly collect any information from anyone, including children under the age of 13. Since we collect no data, there is no data to protect under COPPA, GDPR-K, or similar regulations.

## 6. Security

- All local storage uses iOS sandboxing (no other app can access MyLLM's data)

- API keys (if configured for custom servers) are stored in the iOS Keychain, which is encrypted at rest

- Local network communication uses standard HTTP (not HTTPS) because local LLM servers typically do not support TLS certificates. This traffic does not leave your local network.

- On-device inference processes all data in-memory with no external transmission

## 7. Your Rights

Since we collect no data, there is no data to access, correct, delete, or port. Your data exists only on your device and is fully under your control.

To delete all app data: delete the MyLLM app from your device. This removes all conversations, memories, workflows, settings, and downloaded models permanently.

## 8. Changes to This Policy

If we update this privacy policy, we will post the new version within the app and update the "Last updated" date above. Since we collect no data, changes would typically relate to new features and their local data handling.

## 9. Contact

If you have questions about this privacy policy:

- Email: myllm@opticell-limited.com

---

*MyLLM is designed from the ground up with a simple privacy principle: your data is yours. We never see it, we never touch it, and we never want to.*

# MyLLM — Terms and Conditions

**Last updated:** March 17, 2026

**Effective date:** March 17, 2026

Please read these Terms and Conditions ("Terms") carefully before using the MyLLM application ("the App"). By downloading, installing, or using MyLLM, you agree to be bound by these Terms.

---

## 1. Acceptance of Terms

By accessing or using MyLLM, you confirm that you:

- Are at least 13 years of age (or the minimum age required in your jurisdiction)

- Have read, understood, and agree to these Terms

- Accept our Privacy Policy

If you do not agree, do not install or use the App.

## 2. Description of Service

MyLLM is a local AI assistant application for iOS that:

- Connects to Large Language Model (LLM) servers on your local network (Ollama, LM Studio, or OpenAI-compatible servers)

- Runs small language models directly on your iPhone or iPad for offline use

- Provides agentic tools (web search, file management, contacts, calendar, photos, and more)

- Offers voice input/output, automation workflows, document chat, and translation

- Includes a remote server management terminal for Ollama servers

MyLLM is a **client application**. We do not provide, host, or operate any AI model or server infrastructure. The App connects to services you configure and operate yourself.

## 3. How Large Language Models Work

### 3.1 What is an LLM?

A Large Language Model (LLM) is a type of artificial intelligence that generates text by predicting the most likely next words based on patterns learned during training on large text datasets. LLMs do not "understand" language the way humans do — they are statistical models that produce plausible-sounding text.

### 3.2 Server-Based Models

In server mode, MyLLM sends your messages to an LLM running on a computer you control (your Mac, PC, or Linux server) on your local network. Popular server software includes:

- **Ollama** — runs open-source models like Qwen 2.5, Llama 3.2, Mistral, and others

- **LM Studio** — a desktop application for running models with a graphical interface

- **Custom servers** — any server implementing the OpenAI-compatible API

The quality, accuracy, speed, and capabilities of responses depend entirely on:

- Which model you choose to run (larger models generally produce better results)

- Your server's hardware (RAM, GPU, CPU)

- The model's training data and known limitations

**You are responsible for selecting, configuring, and operating your own LLM server.** We have no control over or visibility into your server or the models you run.

### 3.3 On-Device Models

In offline mode, MyLLM runs small quantized models directly on your iPhone or iPad using the llama.cpp open-source inference engine with Apple Metal GPU acceleration. Available models include:

- **SmolLM 135M** — 100MB, runs on any iPhone, very limited capabilities

- **Qwen 2.5 0.5B** — 350MB, basic conversational ability

- **TinyLlama 1.1B** — 650MB, best on-device quality, requires iPhone 12+

On-device models are significantly smaller and less capable than server-based models. They are suitable for basic conversation, brainstorming, and simple tasks. They do **not** support tool calling (web search, file access, contacts, etc.) or image analysis.

### 3.4 Model Limitations and Risks

All LLMs — whether running on your server or on-device — have inherent limitations that you must understand:

**Accuracy:** LLMs can and do generate incorrect, misleading, or fabricated information (commonly called "hallucinations"). They may present false claims with high confidence. **Never rely on LLM output for medical, legal, financial, or safety-critical decisions without independent verification.**

**Bias:** LLMs reflect biases present in their training data. Outputs may contain cultural, gender, racial, or other biases. This is a property of the model, not the MyLLM application.

**Consistency:** The same prompt may produce different responses each time. LLMs are non-deterministic by design.

**Knowledge cutoff:** Models have a training data cutoff date and do not know about events after that date. MyLLM mitigates this through tool use (web search, news feeds), but the model's reasoning about recent events may still be limited.

**Context limits:** Models can only process a limited amount of text at once (the "context window"). Very long conversations may lose earlier context. MyLLM includes automatic context management, but information may still be lost in extended sessions.

**Tool use reliability:** When MyLLM's agent mode uses tools (web search, file operations, etc.), the model decides which tools to call and how to interpret results. These decisions are not always correct. The model may:

- Call the wrong tool for a task

- Misinterpret tool results

- Fail to use a tool when it should

- Chain tools in unexpected ways in agent mode

**Content:** LLMs may generate content that is offensive, inappropriate, harmful, or factually wrong. Smaller on-device models are less likely to generate harmful content due to their limited capabilities, but no model is guaranteed safe.

### 3.5 Your Responsibility

You understand and agree that:

- AI-generated content should be treated as suggestions, not facts

- You are responsible for verifying any information before acting on it

- You should not input highly sensitive personal data (passwords, financial details, medical records) into any LLM unless you fully understand the privacy implications of your server and model configuration

- You are responsible for the models you download and run, and for complying with their respective licenses

- Agent mode may perform actions (file writes, shortcut triggers) based on AI decisions — use with appropriate caution

## 4. Open-Source Models and Licensing

MyLLM supports a variety of open-source and open-weight models. Each model has its own license that governs its use:

| Model Family | Typical License | Commercial Use |

|---|---|---|

| Qwen 2.5 | Apache 2.0 / Qwen License | Generally permitted |

| Llama 3.2 | Llama 3 Community License | Permitted with conditions |

| Mistral | Apache 2.0 | Permitted |

| SmolLM | Apache 2.0 | Permitted |

| TinyLlama | Apache 2.0 | Permitted |

| DeepSeek | DeepSeek License | Check specific terms |

| Moondream | Apache 2.0 | Permitted |

**You are responsible for reviewing and complying with the license terms of any model you download or use.** MyLLM does not enforce model licenses — it is a general-purpose client that can connect to any compatible model.

## 5. llama.cpp

MyLLM uses llama.cpp, an open-source C/C++ library for LLM inference, licensed under the MIT License. llama.cpp is compiled from source and included in the application binary. It runs entirely on-device and does not transmit any data externally.

llama.cpp is copyright (c) Georgi Gerganov and contributors. The full MIT License text is available at: https://github.com/ggerganov/llama.cpp/blob/master/LICENSE

## 6. User-Generated Content

All content you create within MyLLM (conversations, memories, workflows, personas) is stored locally on your device. You retain full ownership of your content. We claim no rights to any content you create or process through the App.

Content generated by AI models is not owned by us. Ownership of AI-generated content may depend on the license of the model used and the laws of your jurisdiction.

## 7. Acceptable Use

You agree not to use MyLLM to:

- Generate content that is illegal in your jurisdiction

- Attempt to extract training data from models in violation of their licenses

- Use the server terminal to access servers you do not own or have permission to manage

- Circumvent any access controls or security measures on networks or servers

- Generate spam, malware, phishing content, or other harmful material

## 8. Tool and Device Access

MyLLM requests iOS permissions to access device features (contacts, calendar, photos, microphone, camera, files). These permissions are:

- **Requested only when needed** — the app asks for permission the first time a relevant tool is used

- **Controlled by iOS** — you can revoke any permission at any time in Settings → MyLLM

- **Used locally** — accessed data is processed on-device and sent only to your own LLM server (or processed on-device in offline mode)

- **Not transmitted to us** — we never receive any of your personal data

In Agent Mode, the AI may autonomously decide to access tools you have granted permission for. You can review tool calls in the chat interface and disable Agent Mode at any time.

## 9. Server Terminal

The Server Terminal feature sends API commands directly from your device to your Ollama server. You are responsible for:

- Ensuring you have authorization to manage the server you connect to

- Understanding that commands (pull, delete, copy) modify your server's state

- Any bandwidth or storage costs associated with downloading models to your server

## 10. Availability and Support

MyLLM is provided "as is." We do not guarantee:

- Compatibility with any specific LLM server software or version

- That on-device models will produce useful or accurate output

- Uninterrupted operation of the App

- That all features will work with all models (e.g., tool calling requires compatible models)

We aim to address bugs and issues but provide no formal support SLA.

## 11. Intellectual Property

The MyLLM application, including its source code, design, and documentation, is proprietary and protected by copyright. The llama.cpp dependency is used under the MIT License. All model weights are subject to their respective licenses.

"MyLLM" and associated logos are trademarks of their respective owner.

## 12. Disclaimer of Warranties

TO THE MAXIMUM EXTENT PERMITTED BY APPLICABLE LAW, MYLLM IS PROVIDED "AS IS" AND "AS AVAILABLE" WITHOUT WARRANTIES OF ANY KIND, WHETHER EXPRESS, IMPLIED, OR STATUTORY, INCLUDING BUT NOT LIMITED TO IMPLIED WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, AND NON-INFRINGEMENT.

WE DO NOT WARRANT THAT:

- THE APP WILL MEET YOUR REQUIREMENTS

- AI-GENERATED CONTENT WILL BE ACCURATE, COMPLETE, OR RELIABLE

- THE APP WILL BE ERROR-FREE OR UNINTERRUPTED

- ON-DEVICE MODELS WILL PERFORM ADEQUATELY ON YOUR DEVICE

- ANY DEFECTS WILL BE CORRECTED

## 13. Limitation of Liability

TO THE MAXIMUM EXTENT PERMITTED BY LAW, WE SHALL NOT BE LIABLE FOR ANY INDIRECT, INCIDENTAL, SPECIAL, CONSEQUENTIAL, OR PUNITIVE DAMAGES, INCLUDING BUT NOT LIMITED TO:

- LOSS OF DATA OR CONTENT

- ACTIONS TAKEN BASED ON AI-GENERATED INFORMATION

- UNINTENDED FILE MODIFICATIONS MADE BY THE AI AGENT

- SERVER MODIFICATIONS MADE VIA THE SERVER TERMINAL

- INABILITY TO USE THE APP OR ANY FEATURE

- COSTS OF DOWNLOADING MODELS OR RUNNING SERVER HARDWARE

OUR TOTAL LIABILITY SHALL NOT EXCEED THE AMOUNT YOU PAID FOR THE APP.

## 14. Indemnification

You agree to indemnify and hold us harmless from any claims, losses, damages, liabilities, and expenses (including legal fees) arising from:

- Your use of the App

- Your operation of LLM servers

- Content generated by AI models you choose to run

- Your violation of these Terms

- Your violation of any model license

## 15. Termination

You may stop using MyLLM at any time by deleting the App. Upon deletion, all local data (conversations, memories, models, settings) is permanently removed from your device.

We reserve the right to modify or discontinue the App at any time. If we discontinue the App, you retain all data stored locally on your device.

## 16. Changes to These Terms

We may update these Terms from time to time. Changes will be indicated by updating the "Last updated" date. Your continued use of the App after changes constitutes acceptance of the revised Terms. For material changes, we will provide notice within the App.

## 17. Governing Law

These Terms shall be governed by and construed in accordance with the laws of the United Kingdom, without regard to conflict of law provisions. Any disputes shall be resolved in the courts of England and Wales.

## 18. Severability

If any provision of these Terms is found to be unenforceable, the remaining provisions shall continue in full force and effect.

## 19. Entire Agreement

These Terms, together with the Privacy Policy, constitute the entire agreement between you and us regarding MyLLM and supersede any prior agreements.

## 20. Contact

For questions about these Terms:

- Email: myllm@opticell-limited.com

---

*Last reviewed: March 17, 2026*