Most AI apps ask you to trust them. MyBenAI ships a network log so you can see every request it makes—model downloads, searches, connectors—and judge for yourself. This is what data does my AI app send, visible and auditable on your phone.
The Trust Problem
When you use ChatGPT, Claude, or Gemini on your phone, you're trusting the company to tell you what data they collect. They claim "we don't log your messages" or "we only collect aggregated analytics," but you can't verify it. There's no way to open the app and see what's actually being sent over the network.
For on-device AI, the situation is different but still murky. Even if the model runs locally, the app might still make network calls for features like web search, connectors, or telemetry. Without visibility into those calls, you're still trusting the vendor's word.
MyBenAI takes a different approach: transparency through verifiability. The app includes a built-in network log that shows every HTTP request it makes. You can open it, read it, and audit exactly what's leaving your phone.
What a Network Log Is
A network log is a record of every network request the app makes: the URL, the timestamp, the size of the request and response, and the headers. It doesn't show the full request body or response (for privacy), but it shows enough to understand what's happening.
Open the MyBenAI admin console and navigate to the network log. You'll see requests grouped by type: model downloads from Hugging Face, web search requests to DuckDuckGo, connector calls to MCP servers, analytics pings to mybenai.robomiri.com, and error reports to Sentry. Each request is timestamped and labeled with its purpose.
This is verifiable proof. You can watch it in real time, see exactly when requests fire, and confirm what data is being transmitted.
Model Downloads: Transparent and Consented
When you first open MyBenAI, you choose a model. The app downloads it from Hugging Face at runtime—it's not bundled. This network call is visible in the log. You approve the download explicitly; the app shows you the model size and asks for confirmation before starting.
The request to Hugging Face includes only: the model name and repository path. It does not include your chat history, your device ID, or anything about what you intend to do with the model. It's a standard file download, auditable in the network log.
If you switch models later (e.g., upgrade from Qwen3 1.7B to Qwen3.5 4B), another download request appears in the log. You control when this happens.
Optional Web Search: Your Key, Your Control
Web search is opt-in. If you enable it, you provide your own Serper API key or MyBenAI falls back to DuckDuckGo (which requires no key). When you ask a question that triggers a search, the network log shows exactly what was sent.
A search request includes only: your search query and your API key (if using Serper). It does not include your full conversation history, your memory facts, or context unrelated to the search. The query is yours to see and audit.
If you disable web search, zero search requests appear in the log. The feature is genuinely optional.
Connectors: User-Added, User-Audited
Connectors let you extend MyBenAI with remote MCP servers or REST APIs (e.g., Home Assistant, a custom API). You add these yourself—MyBenAI doesn't ship with pre-configured remote connectors.
When you use a connector, the request appears in the network log with the destination URL and timestamp. Because connectors are user-configured, you control what's connected and can inspect the log to confirm requests are going where you expect.
This audit trail is critical. If you accidentally misconfigured a connector, you can catch it immediately and disable it.
Anonymous Usage Analytics: Honest About What's Tracked
Here's where we diverge from "zero telemetry" claims: MyBenAI does send anonymous usage analytics. It's on by default. But it's limited and auditable.
Analytics requests go to mybenai.robomiri.com and include: your install ID (randomly generated, not tied to your Apple/Google account), the device tier (e.g., "flagship" or "mid-range"), the model name you're using, token counts (how many tokens your chats consumed), which personas you've used, and high-level feature flags (e.g., "web search enabled", "RAG active"). It does not include message content, chat transcripts, timestamps of individual chats, or any personally identifiable information.
Why collect this? To understand which models perform well on which devices, how features are used, and where to invest. The same reason a developer might check download counts and crash reports.
You can disable analytics in settings. If you do, no requests to mybenai.robomiri.com appear in the network log.
The honest trade-off: you get a $2 app with no ongoing infrastructure costs and no subscription, but that means no server-side telemetry is needed to justify server costs. We can afford to be minimal and keep everything on your device because we're not running GPU clusters for your inference.
Error Reporting: Sentry for Debugging
If MyBenAI crashes or encounters an error, it sends a report to Sentry (a third-party error-tracking service). This request appears in the network log as a POST to sentry.io.
The report includes: a stack trace (the error's code path), your app version, your device model and OS version, and a sanitized log of recent app activity. It does not include your messages, personal data, or files.
Error reporting helps us fix bugs faster. Like analytics, you can disable it in settings.
What Never Leaves Your Phone
Your message content never leaves your device. This is verifiable: if you search the network log for your actual chat text, you won't find it. Not in any request to any server.
Your chats live in a local SQLite database in your app's private storage. Inference (running the model on your questions) happens entirely on your device. Your data lives on your phone, protected by OS-level encryption while your device is locked.
This is why the network log is so powerful: it proves the absence of message transmission. You can watch your chats load locally, see zero requests containing your words, and have concrete proof.
Inspect for Yourself
The network log is built into the app. Open Settings → Admin Console → Network Requests. Every network call MyBenAI has made since installation is there, sorted by timestamp and destination.
You can filter by URL (e.g., show only Hugging Face requests), export the log as JSON, or share it with a technical friend for review. This is verifiable, auditable privacy—not claims, but data you can examine.
Trust Through Verification, Not Absence
Here's the philosophy: we don't claim zero telemetry (that would be dishonest, given analytics and error reporting). We claim verifiable, limited telemetry. You can see it. You can disable it. You can confirm your messages never appear in it.
This is stronger than a privacy policy because you're not trusting our words—you're reading the network log with your own eyes. If we ever changed our behavior, the log would show it immediately.
MyBenAI's network log is a transparency tool. Use it to audit the app, confirm it matches our claims, and decide if you're comfortable with what leaves your phone. That decision is yours to make based on evidence, not marketing.
Ready to verify what your AI app sends? Get MyBenAI and check the network log yourself. For a deeper dive into on-device architecture, read What Is On-Device AI? or explore how your AI data actually lives on your phone.