Workflow Simulation Engine — Synthetic Process Modeling and Intervention Testing.
Unlike process mining (which harvests actual event logs and analyzes timestamps from real systems), ProSim estimates expected process delays from a parameterized model, then lets you test interventions and measure ROI. Generate, simulate, and optimize business process workflows from natural language descriptions.
- Domain Parsing: Describe a process in natural language, get a BPMN-style workflow graph via Claude API
- Parameterized Graphs: Full parameter schema per node (time, cost, error rate, capacity, parallelization, queue delay)
- Deterministic Simulation: Expected-value computation via probability-weighted DAG traversal
- Monte Carlo Simulation: Stochastic simulation of 10K-1M transactions with configurable distributions
- Bottleneck Detection: Identify the highest-friction nodes in your workflow
- Sensitivity Analysis: Compute marginal impact of each parameter on system-level metrics
- Intervention Layer: Apply what-if modifications, see delta savings and ROI
- Mermaid Diagrams: Export BPMN-style diagrams compatible with any Mermaid renderer
- CLI: Full command-line interface for all operations
- Next.js Dashboard: Modern dark-themed single-page app with live editing and instant re-simulation
- Streamlit Dashboard: Legacy dashboard (still available)
pip install -e ".[dev]"Copy .env.example to .env and set your ANTHROPIC_API_KEY.
cd frontend
npm installOption A: Single command (recommended)
./start.shOption B: Two terminals
# Terminal 1: Start the FastAPI backend
prosim serve --port 8000
# Terminal 2: Start the Next.js frontend (use direct API URL to avoid proxy timeout)
cd frontend
NEXT_PUBLIC_API_BASE=http://127.0.0.1:8000/api npm run devOpen http://localhost:3000. Type a process description, click Generate Baseline, and start tweaking.
Note: Workflow generation takes 30–60 seconds (Claude API). If you see "socket hang up" or ECONNRESET, the Next.js proxy may be timing out. Use NEXT_PUBLIC_API_BASE=http://127.0.0.1:8000/api so the frontend talks directly to the backend.
prosim generate "invoice processing system" -o workflow.json# Deterministic (expected values)
prosim simulate workflow.json --mode deterministic -n 10000
# Monte Carlo (stochastic, 100K transactions)
prosim simulate workflow.json --mode monte_carlo -n 100000 --seed 42
# With sensitivity analysis
prosim simulate workflow.json --sensitivityprosim intervene workflow.json \
--node process_payment \
--time-reduction 30 \
--cost-reduction 20 \
--impl-cost 50000prosim export workflow.json -f mermaid
prosim export workflow.json -f json -o full_export.jsonprosim dashboardProSim/
├── src/prosim/
│ ├── graph/ # Pydantic data models, NetworkX operations, serialization
│ ├── parser/ # Claude API client for NL->workflow generation
│ ├── simulation/ # Deterministic + Monte Carlo engines, bottleneck detection
│ ├── intervention/ # What-if modifications, ROI computation, leverage ranking
│ ├── export/ # Mermaid diagrams, JSON export, Rich reports
│ ├── cli/ # Click CLI commands (generate, simulate, intervene, export, serve)
│ ├── api/ # FastAPI REST layer for the Next.js frontend
│ └── dashboard/ # Legacy Streamlit web UI
├── frontend/ # Next.js 15 + Tailwind CSS + shadcn/ui dashboard
│ └── src/
│ ├── app/ # Next.js App Router (page.tsx, layout.tsx, globals.css)
│ ├── components/ # Dashboard zones (input-bar, diagram, metrics, node-table, whatif, advanced)
│ ├── hooks/ # useProSim state management hook
│ └── lib/ # TypeScript types, API client, utilities
└── tests/ # 103 tests across all subsystems
The Next.js dashboard is a single-page reactive app organized into 5 zones:
- Input Bar — Describe a process or upload JSON
- Diagram + Metrics — Live Mermaid SVG + 4 metric cards + bottleneck list
- Node Table — Click-to-edit parameters, auto re-simulates on change
- What-If — Target a node with time/cost/error reduction sliders, see live delta metrics
- Advanced — Monte Carlo simulation, sensitivity/leverage analysis, export
| Method | Path | Description |
|---|---|---|
| GET | /api/history |
Stub (returns empty list). Accepts ?limit=N. |
| POST | /api/workflow/generate |
Generate workflow from description (Claude API) |
| POST | /api/workflow/parse |
Validate and normalize uploaded JSON |
| POST | /api/simulate |
Run deterministic or Monte Carlo simulation |
| POST | /api/sensitivity |
Sensitivity analysis |
| POST | /api/intervene |
Apply interventions, get before/after comparison |
| POST | /api/leverage |
Rank nodes by improvement leverage |
| POST | /api/export/mermaid |
Generate Mermaid diagram code |
# Python backend tests
pytest tests/ -v
# Build frontend (type checking + production build)
cd frontend && npx next buildWith the backend running (prosim serve --port 8000), run:
./scripts/verify-api.shThis tests the history, parse, and generate endpoints. Generate requires ANTHROPIC_API_KEY.
| Variable | Default | Description |
|---|---|---|
ANTHROPIC_API_KEY |
— | Required for workflow generation |
PROSIM_CORS_ORIGINS |
http://localhost:3000,http://127.0.0.1:3000 |
Comma-separated CORS origins |
NEXT_PUBLIC_API_BASE |
/api |
API base URL (frontend) |
Missing API key: Ensure ANTHROPIC_API_KEY is set in your environment or .env file. Run prosim serve from the project root; it will warn if the key is missing.
Invalid Claude output: If the workflow has orphaned nodes or disconnected graph, the API returns 500 with a message like "Workflow generation produced invalid output". Try again with a clearer process description, or upload a valid workflow JSON instead.
The /api/history endpoint is a stub that returns an empty list. Some browser extensions or dev tools may poll it; the 404 is resolved by the implemented stub.
These warnings appear when Claude returns a workflow with nodes but missing or incorrect edges. The parser attempts to repair (e.g. infer linear chain when edges are empty, fix hyphen/underscore mismatches). If repair fails, generation returns 500. Use the Retry button in the UI or try a different description.
Contributions are welcome! Please read CONTRIBUTING.md for guidelines on how to set up your development environment, run tests, and submit pull requests. By participating in this project you agree to abide by the Code of Conduct.
To report a security vulnerability, please follow the steps in SECURITY.md. Do not open a public issue for security reports.
ProSim is released under the MIT License.