COIN-OR Knowledge Base

Semantically annotated documentation for the COIN-OR open-source optimization stack. Find algorithm details, mathematical formulations, complexity analysis, and academic references — all extracted from annotated C++ headers.

For Students: Getting Started

New to optimization solvers? Here's what you need to dive in.

Browse by Topic

Matrix & Vector Operations

Core sparse data structures used throughout the solver stack.

LU Factorization

Basis matrix factorization for the simplex method.

Presolve & Problem Reduction

Simplify problems before solving to improve performance.

File I/O

Read and write optimization problem formats.

Warm Starting

Save and restore solver state for re-optimization.

Graph Algorithms

Conflict graphs and clique detection for MIP.

Search Trees

Branch-and-bound tree management.

Library Layers

COIN-OR is organized in dependency layers. Higher layers build on lower ones.

Mathematical Formulations

The knowledge base captures mathematical formulations from the solver documentation. For example, the standard form LP:

$$\min_{x} \quad c^T x \quad \text{subject to} \quad Ax = b, \quad x \geq 0$$

And the Markowitz pivot selection criterion minimizes fill-in during LU factorization:

$$\text{score}(a_{ij}) = (r_i - 1)(c_j - 1)$$

where $r_i$ is the row count and $c_j$ is the column count for element $a_{ij}$.

For AI Agents

This knowledge base is built for machine consumption. Point your agent at it and instantly gain expertise on 28 optimization libraries.

Hosted API (No Setup Required)

Fetch the JSON API directly — works with any agent that can read URLs:

https://monistowl.github.io/coin-or-kb/api/annotations.json

Example prompt for Claude, GPT, or any LLM:

Fetch https://monistowl.github.io/coin-or-kb/api/annotations.json and use it to answer questions about COIN-OR optimization libraries. This contains algorithm descriptions, mathematical formulations, and complexity analysis for 2,079 files across 28 libraries (789 with semantic annotations).

API Endpoints:

Local MCP Server (For Claude Desktop)

Add to your Claude Desktop claude_desktop_config.json:

{
  "mcpServers": {
    "coin-or-kb": {
      "command": "python",
      "args": ["/path/to/mcp-server/coin_or_kb_server.py"]
    }
  }
}

Available MCP Tools:

CLI Query Tool

For quick lookups without MCP:

# Search by algorithm
./scripts/kb-query.py algo "Markowitz"          # LU factorization pivot selection
./scripts/kb-query.py algo "branch and bound"   # MIP tree search

# Search by math concept
./scripts/kb-query.py math "clique"             # Conflict graph algorithms
./scripts/kb-query.py math "reduced cost"       # Simplex pricing

# Get library overview
./scripts/kb-query.py lib CoinUtils             # Foundation library
./scripts/kb-query.py lib Ipopt                 # NLP solver

# Get specific file details
./scripts/kb-query.py file CoinUtils CoinFactorization.hpp

# List all documented algorithms
./scripts/kb-query.py list

JSON API

Load the entire knowledge base directly:

import json
with open('site/static/api/annotations.json') as f:
    kb = json.load(f)

# Find all simplex-related algorithms
for layer in kb['layers'].values():
    for lib in layer['libraries'].values():
        for path, file in lib['files'].items():
            algo = file.get('algorithm', '')
            if 'simplex' in algo.lower():
                print(f"{lib['name']}/{path}")
                print(f"  {algo[:100]}...")

API Endpoints:

• /api/annotations.json — Full knowledge base (2.1 MB)
• /api/class-briefs.json — Class index for browser/search (1,963 classes)
• /api/knowledge-graph/ — Concept graph and relationships

What's in the Annotations?

Each annotated file can include:

Coverage: 2,079 annotated files across 28 libraries. 789 files have deep semantic annotations (@algorithm, @math, @complexity).