Docs- Introduction
- Editor
- Philosophy & Vision
- Integration Guide
- Get started
- Syntax
- Variables
- Type system
- Operators
- Control flow
- Functions
- Classes & Inheritance
- Generics
- Enums & When
- Annotations
- Magic const
- Exceptions
- Modules
- Closures
- Standard library
- String
- Array
- Set
- Map
- File
- Date
- Math
- Regex
- JSON
- Runtime / VM
Language Guide
Design Notes
Documentation
Welcome to the official documentation for Autolang — a lightweight execution membrane designed to safely sandbox and execute AI-generated code with near-zero latency and a minimal memory footprint (4.2MB native / 12MB NPM).
Design Philosophy
The "Membrane" Strategy
Autolang does not replace Python or JS ecosystems. It acts as a strict, zero-trust orchestrator where AI intent is decoupled from heavy infrastructure. Expose only the functions you want, and let Autolang handle the safety.
AI-Native Execution
Built for small, dynamic logic scripts generated by LLMs. We prioritize Total Time (compile + run) over deep optimization, achieving 1-3ms cold starts with deterministic opcode fuel limits to prevent infinite loops.
Architecture & VM Status
⚡ Unified Compiler & Runtime
In Autolang, the Virtual Machine (VM) and the Compiler are tightly coupled by design.
Our core priority is maximizing end-to-end execution speed. By unifying the pipeline into a single 4.2MB footprint, we eliminate the heavy cold starts associated with Docker or traditional Node.js isolation environments.
This architecture is specifically tailored for AI Agent workflows. When an LLM generates code, it requires immediate execution and deterministic sandboxing (via Opcode Fuel limits) without the overhead of spinning up mini-operating systems or serializing standalone bytecode.
If you are building Agentic workflows, Multi-Agent systems, or are interested in high-efficiency VM architectures, we'd love to hear from you!
Explore
Community & Contact
Autolang is an open-source project. Contributions, bug reports, and architectural discussions are welcome.
