██████╗ ██████╗ ██████╗ ███████╗
██╔══██╗██╔═══╝ ██╔══██╗██╔════╝
██████╔╝██║ ██║ ██║█████╗
██╔═══╝ ██║ ██║ ██║██╔══╝
██║ ╚██████╗██████╔╝███████╗
╚═╝ ╚═════╝╚═════╝ ╚══════╝
Fluid Language Universal eXecution
A self-assembling, self-improving runtime that compiles markdown to bytecode.
pip install flux-runtime ·
GitHub ·
Playground
pip install flux-runtimeflux hello # Run the hello world demo
flux compile examples/02_polyglot.md -o output.bin # Compile FLUX.MD to bytecode
flux run output.bin # Execute in the VMThat's it. Three commands from zero to running bytecode.
FLUX is a markdown-to-bytecode runtime designed for AI agents. You write structured markdown files containing polyglot code blocks — mixing C, Python, Rust, or any language line by line — and the FLUX compiler weaves them into a single optimized, verifiable bytecode that runs on a 64-register Micro-VM.
Unlike traditional compilers, FLUX treats agents as first-class citizens: the system profiles itself, discovers hot patterns, recompiles bottleneck modules to faster languages, and evolves — all while running.
Think of it as going from orchestra (fixed score) → folk (changes nightly) → jazz (improvises) → rock (pushes limits) → DJ/rave (layers, adapts, self-improves in real-time).
┌─────────────────────────────────────────────────────────┐
│ TIER 8: SYNTHESIS — FluxSynthesizer (the DJ booth) │
│ Wires ALL subsystems together │
├─────────────────────────────────────────────────────────┤
│ TIER 7: MODULES — 8-Level Fractal Hot-Reload │
│ TRAIN → CARRIAGE → LUGGAGE → BAG → ... → CARD │
├─────────────────────┬───────────────────────────────────┤
│ TIER 6A: ADAPTIVE │ TIER 6B: EVOLUTION │
│ Profiler + Selector│ Genome + Mutator + Validator │
├─────────────────────┴───────────────────────────────────┤
│ TIER 5: TILES — 35 composable computation patterns │
├─────────────────────────────────────────────────────────┤
│ TIER 4: AGENT RUNTIME — Trust, scheduling, resources │
├─────────────────────────────────────────────────────────┤
│ TIER 3: A2A PROTOCOL — TELL, ASK, DELEGATE, BROADCAST│
├─────────────────────────────────────────────────────────┤
│ TIER 2: SUPPORT — Optimizer, JIT, Types, Stdlib, Sec │
├─────────────────────────────────────────────────────────┤
│ TIER 1: CORE — FLUX.MD → FIR (SSA) → Bytecode → VM │
└─────────────────────────────────────────────────────────┘
Zero external dependencies — runs on Python 3.10+ stdlib alone.
32 native bytecode instructions for agent-to-agent communication. Agents use TELL, ASK, DELEGATE, and BROADCAST opcodes to coordinate — with trust gating, capability-based routing, and binary serialization.
Write in any language, mix freely, compile to a single binary. C, Python, Rust, TypeScript — they all compile to the same FIR (SSA IR) intermediate representation, then to a unified bytecode.
The universal pivot point. All frontends produce FIR; all backends consume it. 15 types, 42 instructions, SSA form with proper dominators and terminators.
35 reusable, composable computation patterns across 6 categories: COMPUTE, MEMORY, CONTROL, A2A, EFFECT, TRANSFORM. Chain, parallel, and nest tiles to build complex programs from simple pieces.
| # | Example | Description |
|---|---|---|
| 1 | 01_hello_world.py |
3 ways to run FLUX: raw bytecode, FIR builder, full pipeline |
| 2 | 02_polyglot.py |
Mix C + Python in one file |
| 3 | 03_a2a_agents.py |
Agent-to-agent communication |
| 4 | 04_adaptive_profiling.py |
Heat maps & language selection |
| 5 | 05_tile_composition.py |
Composable computation patterns |
| 6 | 06_evolution.py |
Self-improvement engine |
| 7 | 07_full_synthesis.py |
The grand tour — everything wired together |
flux hello Run the hello world demo
flux compile <input> -o <output> Compile source to FLUX bytecode
flux run <bytecode> [--cycles N] Execute bytecode in the VM
flux test Run the full test suite (1848 tests)
flux version Print version info
flux demo Run the synthesis demo
flux info Show system architecture info
flux replay <bytecode> [--verbose] Replay a bytecode trace
flux migrate <path> [--output-dir DIR] Migrate source to FLUX.MD format
flux playground Open the HTML playground
# Compile a C file
flux compile math.c -o math.bin
# Compile a FLUX.MD document
flux compile pipeline.md -o pipeline.bin --verbose
# Run bytecode with a cycle budget
flux run pipeline.bin --cycles 500000
# Migrate an existing Python project
flux migrate src/ --output-dir ./flux_output --verbose
# Migrate a single file
flux migrate calculator.py --lang pythonBring your existing code to FLUX in one command:
# Migrate a Python project
flux migrate my_project/ --output-dir ./flux_output
# Migrate a single C file
flux migrate renderer.c --lang c
# Migrate a directory with verbose output
flux migrate src/ --lang auto --verboseThe migrator produces structured FLUX.MD files with:
## module:header with filename## lang:language identifier### Function:/### Class:/### Struct:sections for each discovered symbol- Original source preserved in code blocks
- FIR IR mapping comments showing how constructs map to FLUX instructions
See tools/flux_migrate.py for the full migration tool with complexity analysis, tile recommendations, and hierarchy mapping.
from flux.pipeline import FluxPipeline
pipeline = FluxPipeline(optimize=True)
result = pipeline.run("""
---
title: My Module
---
## fn: main
```c
int add(int a, int b) {
return a + b;
}""", lang="md")
print(f"Success: {result.success}") print(f"Bytecode: {len(result.bytecode)} bytes")
## For Contributors
1. Fork the repo
2. Create a feature branch (`git checkout -b feature/my-feature`)
3. Write code + tests (`pytest tests/ -v`)
4. Ensure all 1848 tests pass
5. Commit with descriptive message
6. Open a pull request
See [CONTRIBUTING.md](CONTRIBUTING.md) for detailed guidelines.
## Synthesis
FLUX integrates the best ideas from:
| Source | Contribution |
|--------|-------------|
| [nexus-runtime](https://github.com/SuperInstance/nexus-runtime) | Intent-to-bytecode pipeline, A2A opcodes, trust engine |
| [mask-locked-inference-chip](https://github.com/Lucineer/mask-locked-inference-chip) | Zero-software-stack philosophy, hardware-enforced security |
| GraalVM Truffle | Polyglot interop, multi-language type system |
| LLVM | SSA IR, optimization passes |
| WebAssembly | Compact binary, capability security |
| BEAM VM (Erlang) | Zero-downtime hot code reload |
## License
MIT