Felex

Feed ration optimizer for livestock professionals
Free · Offline · Open Source

Download · Installation · Русская версия · Benchmarks

---

What is Felex?

Felex helps livestock specialists put together cost-optimal feed rations without breaking nutritional requirements. It runs entirely on your computer — no cloud, no subscriptions, no internet needed.

Under the hood: a Rust-based linear programming solver, a database of 2,200+ feed ingredients, and an optional local AI advisor that can comment on your ration. Supports dairy and beef cattle, swine, and poultry.

What it does

LP Optimization	Built-in pure-Rust solver (minilp) — no CPLEX, no Gurobi, no license fees
3 Workflows	Build a ration from scratch, complete a partial one, or balance a fixed set of feeds
Smart Constraints	Hard limits + 3 relaxation tiers so the solver finds a solution more often
2,200+ Feeds	Roughages, concentrates, silages, minerals, compound feeds, and more
Multiple Norm Systems	NASEM, INRA, and local Russian standards
Alternative Rations	2–3 variants with different feed mixes and cost/quality comparison
AI Advisor	Local LLM via Ollama (Qwen 3.5) for metabolic commentary and suggestions
Bilingual	Full English and Russian UI — no hardcoded strings
Reports	Export to PDF, Excel, CSV
Fully Offline	Once installed, no internet required

---

How it works

Layer	Tech	Where
Desktop	Tauri 2.0	src-tauri/
API Server	Rust, Axum 0.7, SQLite	src/
Optimizer	good_lp + minilp (pure Rust)	src/diet_engine/
UI	React 18, TypeScript, Tailwind CSS	frontend/src/
State	Zustand + TanStack Query	frontend/src/stores/
Data Pipeline	Python (web scraping, translation)	database/
AI Agent	Ollama / OpenAI-compatible	src/agent/

---

Optimization workflows

1. Build from scratch

Pick an animal type and stage — Felex selects suitable starter feeds from the library and optimizes proportions automatically.

2. Complete a partial ration

You bring one or two feeds you already have on the farm. Felex fills in the missing nutritional roles and balances everything together.

3. Optimize selected feeds only

Felex works strictly with the feeds you picked — no library additions, just proportion tuning.

  Your input ──► Workflow ──► Feed selection ──► LP Solver ──► Alternatives ──► Result
                    │                                               │
                    ├─ From scratch ──► full auto-populate          │
                    ├─ Complete ──────► add only what's missing     ├─► 2–3 variants
                    └─ Selected only ─► keep current set            └─► cost comparison

After solving, the alternatives engine produces 2–3 rations at similar nutritional quality but with different feed combinations — handy when some ingredients aren't available.

---

Benchmark results

Tested across 23 scenarios, 69 runs (v2.0, March 2026):

Metric	Value
Average solve time	213.8 ms
Hard constraint pass rate	64.8%
Norm coverage	81.8%
Average ration cost	183 RUB/day

By workflow

Workflow	Runs	Time (ms)	Pass rate	Coverage	Cost (RUB/day)
From scratch	23	534	66.0%	83.2%	165
Complete	23	107	78.3%	85.0%	248
Selected only	23	0.4	50.2%	77.1%	136

Full methodology and raw data: docs/BENCHMARKS.md

---

Quick start

Ready-made installer (Windows)

Grab the .msi or .exe from the Releases page and run it. Done.

Build from source

You'll need: Node.js 18+, Rust 1.70+, Visual Studio Build Tools (C++ workload)

git clone https://github.com/danilkotelnikov/Felex.git
cd Felex

# Automated setup (Windows)
powershell -ExecutionPolicy Bypass -File scripts/setup.ps1

# Or manually
npm install
npm run build:feed-runtime    # generate feed database artifacts

# Development
npm run dev:full               # starts Rust API + Vite dev server

# Build installer
npm run tauri:build            # produces .msi and .exe

Step-by-step guide with troubleshooting: INSTALL.md

AI advisor (optional)

To get AI commentary on your rations, install Ollama and pull a model:

ollama pull qwen3.5:4b         # light, works on most PCs
ollama pull qwen3.5:9b         # better quality, needs 8+ GB RAM

---

Feed database

Ships with 2,200+ feed ingredients across 9 categories:

Category	Count	Examples
Roughages	1,283	Hay, straw, haylage
Green feeds	411	Pasture grasses, fresh forage
Concentrates	301	Grain, protein meals
Compound feeds	205	Complete feeds, starters
Animal-origin feeds	29	Milk, meat-and-bone meal
Industrial byproducts	28	Oilseed cakes, bran
Mineral supplements	26	Chalk, salts, premixes
Nitrogen supplements	4	Urea

Each entry carries composition data for: dry matter, metabolizable energy, crude protein, amino acids (lysine, methionine+cystine), macro/micro minerals (Ca, P, Mg, Cu, Fe, Zn), vitamins (D3, E, carotene), and crude fiber.

---

Project layout

Felex/
├── src/                        # Rust backend
│   ├── api/                    # HTTP routes (Axum)
│   ├── diet_engine/            # Optimizer, auto-populate, alternatives
│   ├── db/                     # SQLite schema, migrations, models
│   ├── norms/                  # Nutrition norm resolution
│   └── agent/                  # LLM integration
├── frontend/src/               # React + TypeScript UI
│   ├── components/             # UI components
│   ├── stores/                 # Zustand stores
│   ├── lib/                    # API clients, helpers
│   └── generated/              # Auto-generated feed data
├── database/                   # Python data pipeline
│   ├── output/                 # Feed database (JSON, source of truth)
│   └── tests/                  # Pipeline tests
├── src-tauri/                  # Tauri config
├── scripts/                    # Build & setup scripts
└── docs/                       # Documentation, benchmarks

---

Tech stack

Component	Technology
Backend	Rust 2021, Axum 0.7, Tokio
Database	SQLite (rusqlite, bundled)
LP Solver	good_lp 1.7 + minilp (pure Rust)
Desktop	Tauri 2.0
Frontend	React 18.3, TypeScript 5.5
Styling	Tailwind CSS 3.4
Bundler	Vite 5.3
State	Zustand 4.5, TanStack Query 5
UI	Radix UI
i18n	i18next (EN/RU)
Charts	Recharts
Reports	printpdf, rust_xlsxwriter
AI	Ollama (Qwen 3.5 default)
Data pipeline	Python 3 (aiohttp, BeautifulSoup, Pydantic)

---

API

Backend serves REST on localhost:7432/api/v1:

Endpoint	Method	What it does
/feeds	GET	List and filter feeds
/feeds/:id	GET	Single feed with full composition
/rations	GET/POST	List or create rations
/rations/:id	GET/PUT/DELETE	Read, update, delete a ration
/rations/:id/optimize	POST	Run the LP optimizer
/rations/:id/alternatives	POST	Get alternative ration variants
/rations/:id/auto-populate	POST	Preview starter feed plan
/norms	GET	Nutrition norms
/animals	GET	Supported species and stages
/workspace	GET/PUT	User workspace settings
/agent/chat	POST	AI chat (SSE stream)

---

Development

# Backend
cargo run --bin felex-server    # API on :7432
cargo test                      # tests

# Frontend
npm run dev                     # Vite on :5173

# Both at once
npm run dev:full

# Database migrations
cargo run --bin migrate

# Import feed data
cargo run --bin import-feeds

# Rebuild feed artifacts for the UI
npm run build:feed-runtime

# Lint
npm run lint

---

Contributing

Happy to accept contributions:

1. Fork the repo
2. Create a branch (git checkout -b feature/your-idea)
3. Follow the code style described in AGENTS.md
4. Write tests for what you add
5. Open a Pull Request

---

License

MIT — see LICENSE.

---

Rust + React + Tauri
_{For livestock professionals who want reliable ration optimization without cloud dependencies}