ALI separates autonomous agents into 5 components: Causal Core (viability), Ego (behaviour generation), Super-Ego (normative evaluation), Memory (immutable decision records), Runtime (coordination). Zero dependencies. 72 tests. 44 verified compliance requirements. Full plugin system.
ALI Reference Implementation
Artificial Local Intelligence (ALI) — Reference Implementation v0.4
This repository presents an open-source reference architecture and executable Python implementation for Artificial Local Intelligence (ALI), a system designed for enhancing the capabilities of autonomous agents. The architecture distinctly separates five key organizational responsibilities, encapsulating them in independent components:
- Causal Core: Assesses operational viability prior to planning.
- Ego: Generates behavioral proposals devoid of evaluation.
- Super-Ego: Evaluates proposals while refraining from generation.
- Memory: Ensures every decision is recorded as a complete, immutable Event.
- Runtime: Manages the operational cycle without engaging in reasoning.
This implementation adheres to a compliance specification of 44 verifiable SHALL requirements, fulfilling each of them.
Requirements
- Python 3.11 or later
- No third-party packages required
Quick Start
To interact with the system, the following commands can be executed:
# Create demonstration files
python main.py init
# Execute a single cycle (dry run — no files modified)
python main.py run
# Apply approved actions
python main.py run --apply
# Display the current viability state and recent Events
python main.py status
# Reverse the last executed action
python main.py rollback
# Run the complete test suite
python -m unittest discover -s tests -v
Project Structure
The project is organized into several directories and files:
ali/ Core architecture
causal_core.py Operational viability assessment
ego.py Behavioural proposal generation
super_ego.py Normative evaluation
memory.py Immutable Event storage (SQLite)
runtime.py Cycle coordination
interfaces.py Abstract base classes for all components
plugins.py ComponentFactory — plugin system
models.py Architectural data objects (frozen dataclasses)
configuration.py Configuration loading
examples/
llm_ego.py Plugin example: LLM-based Ego (stub mode)
README.md Plugin development guide
config/
ali_config.json Standard configuration
ali_config_llm_example.json Example: LLM Ego via plugin
tests/
test_architecture.py Hard norm and component boundary tests
test_runtime.py Operational cycle and learning tests
test_plugins.py Plugin system tests
test_compliance.py All 44 SHALL requirements verified
workspace/ Default operational domain (local files)
main.py Command-line interface
Plugin System
The architecture allows for seamless replacement of any component without altering the existing codebase. To add a new component, include a components entry in ali_config.json:
{
"components": {
"ego": {
"class": "examples.llm_ego.LLMEgo",
"params": {
"model": "claude-sonnet-4-6",
"stub": true
}
}
}
}
For comprehensive instructions on developing plugins, refer to examples/README.md.
Test Suite
The implementation includes a robust test suite that verifies compliance and functionality:
72 tests — 0 failures
test_compliance.py 41 tests — all 44 SHALL requirements verified
test_plugins.py 15 tests — plugin loading and integration
test_architecture.py 9 tests — component boundaries and hard norms
test_runtime.py 7 tests — operational cycle and learning
References
For further reading on the architectural specifications, consult the following resources:
- Full architectural specification (book):
Stegemann, W. (2026). Artificial Local Intelligence: Architecture and Reference Implementation. Zenodo. Link - Scientific paper:
Stegemann, W. (2026). Artificial Local Intelligence: Architecture and Reference Implementation. Zenodo. Link
The ALI reference implementation provides a structured approach to developing autonomous agents, ensuring modularity and adherence to well-defined operational norms.
No comments yet.
Sign in to be the first to comment.