Debug & Feedback

ETOS LLM Studio puts more thought into "how do I figure out what went wrong" than most LLM clients. It includes a LAN debug client, app log archive, API traffic analysis, and a PoW-protected ticketed feedback system. This page covers what each is and when to reach for it.

Problems You're Likely to Hit

Symptom	Reach for
AI reply didn't appear, error popup	App Logs for the exact error code
AI reply is nonsense	API Traffic Analysis to see the actual request and response
Want to file a bug	Feedback Helper
Want a computer to capture full request traces	LAN Debug
Want to inject a custom proxy / man-in-the-middle	Local Debug Server Proxy

Read This First

App Logs (most-used)

Where

Settings → Extended Capabilities → Extended Features → App Logs

Or tap "View Logs" from an error popup.

What It Records

Daily folders named by date contain structured (JSON) logs of every request and response:

• Each LLM request's full URL, headers, body
• Each response (including streaming chunks)
• Tool call requests and returns
• Failure error codes / stack traces

"Record Plaintext Request Messages"

State	Behavior
Off (default)	Request body logs redact message, content, etc. (structure visible, content hidden)
On	Records plaintext messages, but image, audio, and file Base64 stays redacted (avoids log explosion)

Plaintext Logs Contain Sensitive Content

Turning on plaintext logging means everything you chat about lands on disk. If you don't want that record kept, leave it off and only flip it on during active debugging.

How to Use

1. Reproduce the problem
2. Open App Logs, find today's folder
3. Locate the entry by timestamp
4. Read the error details

Cleanup

• Single day: Swipe to Delete
• "Clear All" wipes history (with confirmation)

Feedback System

Where

Settings → Extended Capabilities → Extended Features → Feedback Helper

Or "Send Feedback" from certain error popups.

Filing Feedback

Each ticket includes:

• Type: Bug / Feature Request / Question / Other
• Title + Body
• Environment info (auto-captured): version, iOS version, device model, simulator status
• Screenshots (optional)
• Log attachment (optional): a redacted log zip

PoW Anti-Spam

Submission requires a Proof of Work — your device spends a few CPU seconds on a hash challenge. This is ETOS's spam guard: no email verification, but every ticket pays a small CPU cost.

Ticket Lifecycle

Submitted tickets get a number. The feedback list shows:

• Status: Unread / Read by Developer / In Progress / Resolved / Closed
• Comment thread: developers reply on the ticket; you reply back
• Developer flags: "Confirmed Bug", "Cannot Reproduce", "Fixed in Next Version", etc.

State auto-syncs.

Two-way Tickets

• iPhone tickets sync to the Watch
• System notifications push status updates

API Traffic Analysis

Where

Settings → Extended Capabilities → Extended Features → Advanced Diagnostics

Simple Mode (token usage only)

For just token usage and response speed, Usage Analytics is enough.

Full Traffic Capture

Advanced Diagnostics captures the complete flow per request:

• Full URL (including query string)
• Full request headers (including Authorization)
• Full request body
• Full response headers
• Full response body (streaming-merged)

Best for figuring out "why did the AI say what it said."

Advanced

LAN Debug Mode

Top of Advanced Diagnostics, Connection Mode: choose HTTP Polling or WebSocket.

Use with the PC-side debug tools in docs/debug-tools and docs/debug-tools-go.

How It Works

The device proactively connects to the PC-side WebSocket server, receives commands, and executes file operations.

The device dials out — no need to open router ports. The PC runs a WebSocket server (default port 8765); the device connects and lets you:

• Browse the app's sandbox files
• Pull the local database for diagnostics
• Upload / download files
• Watch a live console in a browser

Steps

1. PC: run the server in docs/debug-tools/ (npm project)
2. PC server prints its local IP + port when started
3. iPhone: Advanced Diagnostics → enter "IP:port" (e.g. 192.168.1.100:8765) → tap Connect
4. When green, watch the live log in your browser

Only on Trusted Networks

Advanced Diagnostics fully exposes the app sandbox to the PC. Don't use it on coffee-shop Wi-Fi or any public network.

The on-page warnings:

Use only on trusted networks Disconnect promptly after use

API Proxy Mode

The API Proxy Settings option routes all LLM requests through the PC-side man-in-the-middle:

Set API Base URL to: http://PC-IP:8080

Requests get logged, parsed, optionally modified, then forwarded to the real LLM API. Good for:

• Inspecting full streaming response content
• Writing your own middleware to alter requests / responses
• Mocking error responses to test app behavior

Global Proxy (Production Networks)

Settings → Providers & Models → Global Proxy

If every LLM request needs a proxy:

Field	Notes
Enable Proxy	Master switch
Proxy Type	HTTP / HTTPS / SOCKS5
Proxy Host	Hostname or IP
Port	1–65535
Username (optional)	Proxy auth
Password (optional)	Proxy auth

Priority (covered in Add Your First Provider advanced): per-provider proxy > global proxy.

Best for:

• Network-wide proxy requirement (corporate, VPN)
• You don't want to configure each provider separately

Log Redaction Behavior

ETOS redacts by default:

Field	Recorded
URL	✅ Full
Headers	✅ Full (Authorization included — for debugging; redact before sharing)
Body messages text	❌ Hidden by default (turn on "Record Plaintext Messages" to include)
Body image base64	❌ Always hidden
Body audio base64	❌ Always hidden
Body file base64	❌ Always hidden
Streaming response body	✅ Recorded merged

Redact Again Before Sharing

Even with built-in redaction, logs still contain URLs and headers that may include API keys. Before sharing with a developer, do another pass: replace Authorization / X-API-Key etc. with ***.

Model Connectivity Test

To quickly verify whether a newly added model is reachable, without going through chat:

Settings → Providers & Models → \[provider\] → top-right "Model Test" button

It fires a minimal request against every enabled model in that provider and reports:

• ✅ Passed
• ❌ Error message
• Response time

The fastest way to identify "which models still work and which ones broke."

Next

• Everyday small tricks → Hidden Gems
• What the Watch can do → Using Apple Watch
• Stuck on a specific issue → FAQ