title	Optimization Agent
emoji	🎯
colorFrom	blue
colorTo	purple
sdk	streamlit
sdk_version	1.40.0
app_file	src/ui/app.py
pinned	false

OR-Tools & LangGraph Optimization System

Multi-agent orchestration that pairs LangGraph LLM nodes with deterministic Google OR-Tools solving and a Streamlit UI.

Architecture

flowchart LR
    UI[Streamlit UI] --> F[Formulator]
    F --> C[Coder]
    C --> E[Executor]
    E -->|success| R[Reporter]
    E -->|error / infeasible| V[Reviewer]
    V --> C
    R --> UI

Loading

Node	Role
Formulator	Business prompt + baseline params → Pydantic StructuredOptimizationModel
Coder	Fills CP-SAT or VRP templates (no free-form OR-Tools syntax)
Executor	Subprocess sandbox; captures stdout / tracebacks
Reviewer	Self-correction loop on errors or infeasibility (max 3 retries)
Reporter	Human-readable markdown summary

Requirements

• Python 3.12
• uv

Setup

cd optimisation-agent
cp .env.example .env
# Optional: set OPENAI_API_KEY for LLM nodes (heuristic fallback works offline)

uv sync --extra dev

Run UI

uv run streamlit run src/ui/app.py

The UI is a guided 4-step workflow (problem type → parameters → data → Run optimizer).
By default it uses the direct OR-Tools engine (fast, no API key). Enable “AI pipeline” only for natural-language refinements.

Run tests

uv run pytest

Project layout

src/
  agents/          # formulator, coder, reviewer, reporter
  graph/           # OptimizationState + StateGraph builder
  schemas/         # Pydantic models
  solver_sandbox/  # executor + OR-Tools templates
  ui/              # Streamlit app
  utils/           # logging, LLM helpers
tests/

Environment variables

Variable	Description
OPENAI_API_KEY	Enables LLM-backed nodes (optional)
OPENAI_MODEL	Default gpt-4o-mini
MAX_SOLVER_RETRIES	Reviewer loop limit (default 3)