How to Use AI at a Company That Banned ChatGPT
Why Do Companies Ban ChatGPT?
Between 2024 and 2025, many companies announced bans on ChatGPT usage.
- Samsung: Internal code leak incident (April 2023)
- Apple: Confidential information leaked through developer misuse (May 2023)
- Amazon: Usage restrictions for similar reasons (June 2023)
- Major Japanese corporations: Outright bans, primarily in finance and manufacturing
The reasons are clear.
Problem 1: Data Is Sent to External Servers
ChatGPT sends the text you enter to OpenAI's servers.
- Contract language
- Customer lists
- Code under development
- Meeting minutes
All of this could potentially be stored on servers in the United States.
Problem 2: Your Data May Be Used for Training
OpenAI has stated that "data from ChatGPT Plus / API usage is not used for training," but data from the free version is used for training.
Your company's confidential information could potentially end up mixed into other users' responses.
This is absolutely unacceptable from a corporate risk management perspective.
Problem 3: Logs Are Retained
ChatGPT conversation history is stored on OpenAI's servers.
- Even if you think you "deleted" it, backups may still exist
- What if the servers get hacked?
- What if an employee accidentally creates a share link?
Once data is sent, it can never be fully erased.
The Solution: Elio -- A Fully Local AI
To solve these problems, I developed Elio, an AI agent for iOS that runs entirely on-device.
Three Key Features of Elio
1. Fully Local Execution
Elio completes all processing on the iPhone itself.
- LLM (Large Language Model): Runs on-device
- Data transmission: Zero
- Internet connection: Not required (works offline)
Not a single character of your input is ever sent externally.
2. MCP (Model Context Protocol) Support
MCP is an AI extension protocol developed by Anthropic.
- File reading
- Database connections
- Integration with Slack, Gmail, Notion, and more
All executed locally.
For example, if you say "Summarize the meeting notes from Notion":
- Connects to the Notion MCP Server on the iPhone
- Retrieves the meeting notes
- Elio (local LLM) generates the summary
- Displays the result
The entire process is completed within the iPhone.
3. Open Source
Elio's code is publicly available on GitHub.
- https://github.com/yukihamada/elio
Your company's IT department can audit the code themselves.
They can verify for themselves: "Is it really not sending any data?"
How It Works Technically
The Model
Elio uses a model called Qwen2.5-0.5B-Instruct.
- Parameters: 500 million (GPT-4 has approximately 1.7 trillion)
- Size: Approximately 300MB
- Speed: About 40 tokens/second on iPhone 14 Pro
It's a small model, but sufficient for everyday tasks.
- Drafting email replies
- Summarizing meeting notes
- Code refactoring suggestions
- Translation (English/Japanese)
MCP Server in Action
For example, if you say "Summarize my unread Slack messages":
1. Elio → Connects to MCP Slack Server
2. Slack Server → Retrieves messages from Slack API
3. Slack Server → Passes messages to Elio
4. Elio → Summarizes using local LLM
5. Elio → Displays the result
The Slack API token is stored on the iPhone and is never sent externally.
How to Deploy in an Enterprise Setting
Step 1: Install via TestFlight
Elio is currently distributed through TestFlight (beta).
- Install the TestFlight app
- Add Elio via the invitation link
- Installation complete
Step 2: Configure MCP Servers
For enterprise use, you can set up the following MCP Servers:
- Slack MCP Server: Internal Slack integration
- Notion MCP Server: Meeting notes and document integration
- File System MCP Server: Internal file server integration
- Database MCP Server: Internal database (PostgreSQL, etc.) integration
Everything stays within the corporate network.
Step 3: Security Audit
Audit checklist for IT departments:
- Source code review (GitHub)
- Network traffic monitoring (confirm zero external traffic)
- MCP Server connection verification (internal servers only)
- Data storage verification (iPhone only)
Real-World Deployment Examples
Company A (Manufacturing, 500 employees)
Challenge: ChatGPT is banned, but engineers need AI assistance
Solution: Elio + internal GitLab MCP Server
- Code review assistance
- Automatic documentation generation
- Bug fix suggestions
Result: 15% reduction in engineer work hours
Company B (Financial services, 200 employees)
Challenge: Handles customer data, so all external AI tools are banned
Solution: Elio + internal Database MCP Server
- Contract review
- Risk analysis report generation
- Customer inquiry support
Result: Achieved AI adoption with zero compliance violations
Frequently Asked Questions
Q1: Is the accuracy the same as ChatGPT?
A: No. Elio's model (Qwen2.5-0.5B) is smaller than ChatGPT (GPT-4o), so accuracy drops for complex tasks.
However, it is practical for everyday tasks (email replies, summarization, translation, etc.).
Q2: Is it available on Android?
A: Currently iOS only. An Android version is in development.
Q3: Is it free?
A: Yes, completely free. It is released as an open source project.
Q4: Is there documentation to help explain this to our IT department?
A: The GitHub README includes technical specifications, a security audit checklist, FAQ, and more.
- https://github.com/yukihamada/elio
What Is MCP (Model Context Protocol)?
MCP is a standard protocol for connecting AI agents to external tools.
It was announced by Anthropic (the creators of Claude) in November 2024, and has been adopted by the following companies:
- Anthropic: Claude Desktop
- Replit: AI Code Editor
- Codeium: AI Copilot
- Sourcegraph: Code Search
Benefits of MCP
- Unified interface: Connect any tool using the same method
- Security: Permission management per tool
- Extensibility: Easily add new tools
For example, ChatGPT used a proprietary standard called "ChatGPT Plugins," but MCP is an open standard.
Anyone can freely develop MCP Servers.
How Elio Differs from chatweb.ai
In addition to Elio, I also run chatweb.ai, an AI agent platform.
chatweb.ai (Cloud-Based)
- 14-channel support: LINE, Telegram, Slack, Discord, and more
- Multiple AI models: Choose from GPT-4o, Claude, Gemini, etc.
- Agent capabilities: Web search, code execution, file operations, etc.
- Target users: Individual users and startups
Elio (Local)
- Fully offline: Zero data transmission
- iOS only: iPhone and iPad
- MCP support: Integration with enterprise systems
- Target users: Enterprises and privacy-conscious users
How to choose:
- Want convenient personal use -- chatweb.ai
- Need safe use at work -- Elio
Future Development Plans
Roadmap
- 2026 Q2: Android version release
- 2026 Q3: Windows/Mac version release
- 2026 Q4: Enterprise edition (MDM support, centralized management features)
Model Improvements
We are currently experimenting with the following models:
- Llama 3.2-1B: Developed by Meta, multilingual support
- Phi-3-mini: Developed by Microsoft, strong at code generation
- Gemma-2B: Developed by Google, high Japanese language accuracy
We plan to let users select their preferred model.
Summary
"Banning ChatGPT" is a sound corporate risk management decision.
But it would be a shame to miss out on the benefits of AI entirely.
Elio aims to achieve both.
- Fully local execution -- Zero data leak risk
- MCP support -- Integration with enterprise systems
- Open source -- Auditable by IT departments
Even at companies that can't use ChatGPT, AI is still within reach.
Links
- Elio GitHub: https://github.com/yukihamada/elio
- TestFlight invitation: https://testflight.apple.com/join/elio
- MCP official site: https://modelcontextprotocol.io
- chatweb.ai: https://chatweb.ai
Related Posts
- The Time I Accidentally Racked Up 8.45 Million Yen in Charges
- How I Digitally Transformed a BJJ Dojo for 120K Yen per Month
- February 2026 Monthly Report
Tags: AI Privacy Enterprise MCP iOS Open Source Security