AI-Powered Android Debug Agent
Turn Android screen recordings into logs, source code insights, and AI-powered root cause analysis.
Android debugging usually looks like this:
Screen Recording
+
Android Export Log
+
Source Code
↓
Watch Video
↓
Remember Timestamp
↓
Extract Logs
↓
Search Logcat
↓
Search Source Code
↓
Analyze Root Cause
A single bug can take 30 minutes to several hours to investigate.
Logcat Agent automates this workflow.
Video
↓
Timestamp OCR
↓
Voice Understanding
↓
Logcat Location
↓
Source Mapping
↓
Flow Analysis
↓
ChatGPT
↓
Root Cause Report
Extract timestamps directly from screen recordings.
The distributed Python service should be built from the runtime-only dependency set:
./scripts/build_python_runtime.shThis runtime intentionally excludes training/CUDA packages:
torch
torchvision
torchaudio
nvidia-*
triton
OCR inference uses onnxruntime and timestamp_crnn_best.onnx; subtitles use the FunASR ONNX runtime. PyTorch is only needed on the training machine to train/export the OCR model.
Put the FFmpeg binaries for each target platform here before building:
third_party/
ffmpeg/
windows/
ffmpeg.exe
ffprobe.exe
macos/
ffmpeg
ffprobe
linux/
ffmpeg
ffprobe
Only ffmpeg and ffprobe are required for the platform you are packaging. The desktop builds copy them next to the app executable, and the app uses the bundled FFmpeg first instead of depending on the user's system environment or PATH.
During development, the local FunASR model can live next to the OCR model:
pythonai/
timestamp_crnn_best.onnx
funasr-paraformer-zh/
config.yaml
model_quant.onnx
tokens.json
...
All runtime models can be installed from one downloadable archive:
./scripts/install_models.sh https://example.com/logagent-models.zipExpected archive layout:
logagent-models/
timestamp_crnn_best.onnx
funasr-paraformer-zh/
config.yaml
model.onnx or model_quant.onnx
tokens.txt or tokens.json
...
At runtime the same archive can be configured with:
export LOGCAT_AGENT_MODEL_BUNDLE_URL=https://example.com/logagent-models.zipSupported formats:
05-07 16:46:59
05-07 16:46:59.170
2026-05-07 16:46:59
Extract bug descriptions from tester recordings.
Example:
Tester:
点击自动泊车以后，界面卡住了，车辆没有进入泊车流程。
Automatically converted to:
{
"bug_description": "点击自动泊车以后界面卡住，车辆没有进入泊车流程"
}Automatically:
Locate Timestamp
↓
Search Related Logs
↓
Group By Tag
↓
Highlight Critical Events
Automatically map:
Tag
↓
Class
↓
Source File
↓
Function
Example:
AutoApa_EngineSomeIpNativeManager
↓
EngineSomeIpNativeManager.cpp
Line 456
Generate Mermaid flowcharts from logs.
flowchart TD
A[AutoApa Init]
B[Request Render Data]
C[Commit Render]
D{Timeout?}
E[Error]
F[Success]
A --> B
B --> C
C --> D
D -->|Yes| E
D -->|No| F
Combine:
Bug Description
+
Timestamp
+
Related Logs
+
Source Code
+
Flowchart
Generate:
Problem Summary
Key Timeline
Related Logs
Related Source Files
Possible Root Cause
Suggested Fix
Flutter Desktop
│
├── Video Player
├── Subtitle Overlay
├── Logcat Viewer
├── Source Viewer
└── AI Debug Tree
│
▼
Rust
│
├── ZIP Extraction
├── tar.lz4 Extraction
└── Audio Extraction
│
▼
Python
│
├── OCR Agent
├── Voice Agent
├── Log Agent
├── Source Agent
├── Flow Agent
└── ChatGPT Agent
│
▼
AI Root Cause Report
Import Video
↓
Import Android Export Log
↓
OCR Timestamp
↓
Voice Transcription
↓
Locate Logcat
↓
Generate Flowchart
↓
Map Source Code
↓
ChatGPT Analysis
↓
Root Cause Report
- Flutter
- media_kit
- flutter_rust_bridge
- Rust
- Python
- ONNXRuntime
- FunASR
- ChatGPT
- ripgrep
- Universal Ctags
- Video OCR
- Android Log Extraction
- Timestamp Location
- Bug Voice Agent
- Flutter Logcat Viewer
- Source Agent
- Flow Agent
- ChatGPT Root Cause Agent
- Android Debug Report Generator
- Autonomous Android Debug Agent
Most tools only analyze logs.
Logcat Agent understands:
Video
+
Voice
+
Logs
+
Source Code
and turns them into actionable debugging insights.
If this project helps you, please give it a Star.