{"slug": "ali-a-python-agent-architecture-where-the-normative-evaluator-is", "title": "ALI: A Python agent architecture where the normative evaluator is", "summary": "Independent researcher Wolfgang Stegemann released v0.4 of the open-source reference implementation of the Artificial Local Intelligence (ALI) architecture, a Python-based agent framework that separates five organizational responsibilities into independent components. The implementation satisfies all 44 verifiable SHALL requirements defined in the architectural specification and requires no third-party packages, running on Python 3.11 or later.", "body_md": "**Artificial Local Intelligence — Reference Implementation v0.4**\n\nThis repository contains the open-source reference implementation of the ALI architecture as described in:\n\nStegemann, W. (2026). Artificial Local Intelligence: Architecture and Reference Implementation. Zenodo.\n\n[https://doi.org/10.5281/zenodo.21631699]\n\nArtificial Local Intelligence (ALI) is a reference architecture for autonomous agents that separates five organisational responsibilities into independent components:\n\n**Causal Core**— evaluates operational viability before planning begins** Ego**— generates behavioural proposals without evaluating them** Super-Ego**— evaluates proposals without generating them** Memory**— preserves every decision as a complete, immutable Event** Runtime**— coordinates the cycle without participating in reasoning\n\nThe architecture is defined by a compliance specification of 44 verifiable SHALL requirements. This implementation satisfies all 44.\n\n- Python 3.11 or later\n- No third-party packages\n\n```\n# Create demonstration files\npython main.py init\n\n# Run one cycle (dry run — no files changed)\npython main.py run\n\n# Execute approved actions\npython main.py run --apply\n\n# Show current viability state and recent Events\npython main.py status\n\n# Undo the last executed action\npython main.py rollback\n\n# Run the full test suite\npython -m unittest discover -s tests -v\nali/                    Core architecture\n    causal_core.py      Operational viability assessment\n    ego.py              Behavioural proposal generation\n    super_ego.py        Normative evaluation\n    memory.py           Immutable Event storage (SQLite)\n    runtime.py          Cycle coordination\n    interfaces.py       Abstract base classes for all components\n    plugins.py          ComponentFactory — plugin system\n    models.py           Architectural data objects (frozen dataclasses)\n    configuration.py    Configuration loading\n\nexamples/\n    llm_ego.py          Plugin example: LLM-based Ego (stub mode)\n    README.md           Plugin development guide\n\nconfig/\n    ali_config.json                Standard configuration\n    ali_config_llm_example.json    Example: LLM Ego via plugin\n\ntests/\n    test_architecture.py    Hard norm and component boundary tests\n    test_runtime.py         Operational cycle and learning tests\n    test_plugins.py         Plugin system tests\n    test_compliance.py      All 44 SHALL requirements verified\n\nworkspace/              Default operational domain (local files)\nmain.py                 Command-line interface\n```\n\nAny architectural component can be replaced without touching the rest of\nthe code. Add a `components`\n\nentry to `ali_config.json`\n\n:\n\n```\n{\n  \"components\": {\n    \"ego\": {\n      \"class\": \"examples.llm_ego.LLMEgo\",\n      \"params\": {\n        \"model\": \"claude-sonnet-4-6\",\n        \"stub\": true\n      }\n    }\n  }\n}\n```\n\nSee `examples/README.md`\n\nfor a complete guide to writing plugins.\n\n```\n72 tests — 0 failures\n\ntest_compliance.py    41 tests — all 44 SHALL requirements verified\ntest_plugins.py       15 tests — plugin loading and integration\ntest_architecture.py   9 tests — component boundaries and hard norms\ntest_runtime.py        7 tests — operational cycle and learning\n```\n\nFull architectural specification (book):\n\nStegemann, W. (2026). Artificial Local Intelligence: Architecture and Reference Implementation.\nZenodo. [https://doi.org/10.5281/zenodo.21631699](https://doi.org/10.5281/zenodo.21631699)\n\nScientific paper:\n\nStegemann, W. (2026). Artificial Local Intelligence: Architecture and Reference Implementation.\nZenodo. [https://doi.org/10.5281/zenodo.21632233](https://doi.org/10.5281/zenodo.21632233)\n\nWolfgang Stegemann\n\nIndependent Researcher\n\nORCID: 0009-0000-1346-1170 [https://orcid.org/0009-0000-1346-1170](https://orcid.org/0009-0000-1346-1170)\n\nMIT License. See LICENSE file.", "url": "https://wpnews.pro/news/ali-a-python-agent-architecture-where-the-normative-evaluator-is", "canonical_source": "https://github.com/drwolfgangstegemann-sudo/ali-reference-implementation", "published_at": "2026-07-28 09:05:04+00:00", "updated_at": "2026-07-28 09:23:27.352380+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-agents", "ai-research", "ai-tools", "developer-tools"], "entities": ["Wolfgang Stegemann", "Artificial Local Intelligence", "ALI", "Python", "Zenodo"], "alternates": {"html": "https://wpnews.pro/news/ali-a-python-agent-architecture-where-the-normative-evaluator-is", "markdown": "https://wpnews.pro/news/ali-a-python-agent-architecture-where-the-normative-evaluator-is.md", "text": "https://wpnews.pro/news/ali-a-python-agent-architecture-where-the-normative-evaluator-is.txt", "jsonld": "https://wpnews.pro/news/ali-a-python-agent-architecture-where-the-normative-evaluator-is.jsonld"}}