QuakeCore AI Agent

QuakeCore is an AI-based seismic data processing agent framework. It allows users to upload various seismic data formats (MiniSEED, SAC, SEG-Y, HDF5) and interact with an AI via natural language to analyze file structures, get statistics, and perform phase picking.

Features

• Multi-Format Support: Reads SEGY, MiniSEED, SAC, HDF5, NumPy arrays.
• Smart Phase Picking: Built-in STA/LTA, AIC, and other traditional picking algorithms.
• API Backend: FastAPI routes for chat, uploads, config, skills, and artifacts.
• Frontend: A ChatGPT-like Next.js chat UI with integrated click upload, drag-and-drop upload, and paste upload.
• Local/Cloud AI Support: Integrates with local Ollama or cloud-based DeepSeek APIs.

Quick Start

1. Prerequisites

• Conda — for managing the Python environment (Miniforge recommended, or Miniconda / Anaconda)
• Node.js & npm — for running the frontend (Download)

2. Installation

git clone https://github.com/Chuan1937/QuakeCore.git
cd QuakeCore

# Create and activate a conda environment
conda create -n quakecore python=3.12 -y
conda activate quakecore

# Install dependencies
pip install -r requirements.txt

# Install backend API dependencies
pip install -r requirements-backend.txt

2. Configure LLM

QuakeCore supports two providers:

• DeepSeek API
• Ollama

Recommended DeepSeek setup:

export DEEPSEEK_API_KEY=your_key

Recommended defaults:

• Provider: deepseek
• Model: deepseek-v4-flash
• Base URL: https://api.deepseek.com

Ollama example:

ollama pull qwen2.5:3b

Default Ollama base URL:

• http://localhost:11434

3. Start the Backend API

conda activate quakecore
uvicorn backend.main:app --host 127.0.0.1 --port 8000 --reload

Backend URLs:

• API: http://127.0.0.1:8000
• OpenAPI docs: http://127.0.0.1:8000/docs

4. Start the Frontend

cd frontend
npm install
npm run dev

Open:

• Frontend: http://localhost:3000

If the frontend should talk to a non-default backend, set:

export NEXT_PUBLIC_API_BASE_URL=http://127.0.0.1:8000

By default, the frontend uses http://127.0.0.1:8000.

Configuration

Frontend Settings Page

After starting backend and frontend, open the settings page in the web UI and save model settings there.

• DeepSeek:
  • Provider: deepseek
  • Model: deepseek-v4-flash
  • API Key: can be left empty if DEEPSEEK_API_KEY is already exported
  • Base URL: https://api.deepseek.com
• Ollama:
  • Provider: ollama
  • Base URL: usually http://localhost:11434
  • Model: detected from the local Ollama server

Usage

1. Configure your LLM settings in the sidebar.
2. Upload a seismic data file (or use the examples in example_data/).
3. Chat with the AI! Try prompts like:
   • "Analyze this SEGY file's structure."
   • "What is the sampling rate of this file?"
   • "Perform phase picking on the loaded waveform."
   • "Convert this data to HDF5 format."
4. In the new frontend chat page, you can upload files with:
   • Click upload (+ button in composer)
   • Drag-and-drop to the chat page
   • Paste file(s) from clipboard
5. After upload, trigger capabilities with natural language directly:
   • "please do phase picking"
   • "请分析当前文件结构"
   • "对当前波形做初至拾取"
   • "使用当前数据进行地震定位"
   • "帮我做连续地震监测"
   • "对加州2019年7月4日的17到18点进行地震监测"

Example: Phase picking results displayed in the chat interface.