A locally-deployable supply chain risk analysis tool for supply chain managers. Import Excel/CSV, get risk insights, reasoning trails, and actionable recommendations — all offline, no cloud dependency.
| Output | Answers the Question |
|---|---|
| Risk Trend Chart | |
| Is risk worsening or improving over time? | |
| Risk Distribution Chart | |
| Where is risk concentrated across the supply chain? | |
| Propagation Timeline | |
| Where does risk come from and spread to? | |
| High-Risk Node Table | |
| Which nodes need immediate action, in what order? | |
| Data Confidence | |
| How reliable is this analysis? What data is missing? | |
| Domain Insights | |
| Bullwhip effect, VMI, QR patterns detected? |
Each high-risk node includes a full 6-step reasoning trail, risk cause details (trigger metrics + threshold comparisons), and structured action recommendations (replenish / reroute / switch supplier / adjust logistics / investigate / monitor).
Download the latest .dmg
from GitHub Releases → drag to Applications → launch Saisca. (First launch: right-click → Open.)
Now free and open — no activation code required.
git clone https://github.com/cayincoorts-hue/saisika.git
cd saisika
pip install -r requirements.txt
python run.py
demo_data/demo_scenario/
contains a purpose-built 10-node supply chain simulation (26 weeks × 5 metrics):
| File | Description |
|---|---|
Sales Order.csv |
|
| Downstream demand (long format: date, node_id, value) | |
Production.csv |
|
| Factory output | |
Delivery To Distributor.csv |
|
| Distribution center deliveries | |
Factory Issue.csv |
|
| Supplier shipments | |
Inventory.csv |
|
| Node inventory levels (S002 sharp drop in last 10 weeks) | |
Nodes.csv |
|
| Node attributes (name, type, tier, region) | |
Edges.csv |
|
| Supply chain relationships (risk links marked) | |
Node Types.csv |
|
| Node type descriptions | |
README.md |
|
| Scenario design notes |
Engineered risk characteristics:
- S002 East China Parts Supplier → inventory crash + delivery disruption → high-risk node - P001 Shenzhen Factory → upstream volatility amplification → bullwhip effect - D001 South China DC → volatility significantly lower than peers → VMI pattern - R002 Shanghai Retailer → high-frequency small-batch stable demand → QR pattern
User uploads Excel/CSV
↓
excel_adapter Read file → detect role (fact/node/edge/metadata)
↓
field_mapper Wide-table melt → column mapping → tri-state identification
↓
data_merger Cross-scenario merge → mark duplicate measurement bases
↓
graph_builder Build supply chain network → infer tier hierarchy
↓
risk_engine Compute risk scores → annotate risk_causes
↓
decision_engine Classify actions → generate action_type + justification
↓
analysis_engine Assemble 6 result objects (including domain_insights)
↓
prompt_builder Generate text conclusions (reads annotations, never raw numbers)
↓
result_exporter Output JSON + HTML report
| Layer | Technology |
|---|---|
| Frontend | React + TypeScript + Vite + ECharts + GSAP |
| Backend | Python + FastAPI + pandas + openpyxl |
| Graph Layout | Three.js + react-force-graph-3d |
| LLM (optional) | Ollama (explanation layer only, not risk engine) |
| Packaging | PyInstaller + Electron + electron-builder |
| Theory | SCOR / Bullwhip Effect / VMI / QR / SC-BSC |
The system accepts two fact-table formats:
Long format (recommended): date, node_id, value
date,node_id,value
2026-01-05,S001,99.3
Wide format: rows as dates, columns as node codes
Date,SOS008L02P,SOS005L04P
2026-01-05,1355.0,890.2
Node tables need: node_id, node_name, node_type
. Edge tables need: source, target
.
- All data stored locally in
data/
directory - No internet connection, no telemetry, no cloud dependency
- Backend compiled to binary for code logic protection
MIT License. See LICENSE for details.
中文说明 #
面向供应链管理者，可本地部署、支持 Excel/CSV 导入的风险分析桌面工具。不依赖云端，数据不出本地。
| 结果 | 回答的问题 |
|---|---|
| 风险趋势图 | 风险是在加剧还是缓解？ |
| 风险分布图 | 整体风险集中在哪一层？ |
| 风险传播时序图 | 风险从哪里来、往哪里扩散？ |
| 高风险节点表 | 哪些节点需要立即处理、按什么顺序？ |
| 数据可信度 | 结论有多可信、缺了什么数据？ |
| 供应链领域洞察 | 是否存在牛鞭效应、VMI、QR 等经典模式？ |
从 GitHub Releases 下载 .dmg
→ 拖入 Applications → 右键打开。
现已免费开放，无需激活码。
demo_data/demo_scenario/
含 10 节点供应链模拟数据（26 周 × 5 指标），刻意设计了牛鞭效应、VMI、QR 等经典风险特征。
全部数据存本地 data/
目录。不联网，无遥测，无云端依赖。
MIT License。详见 LICENSE。