A governed cognitive substrate for AI systems.
Models propose. Alethic decides what may enter state and what may become action.
Alethic places a small, enforceable kernel between reasoning components and external effects. It maintains typed state, links beliefs to evidence, applies declarative constraints, and records why each proposal was committed or invalidated. The kernel contains no model, prompt, tool, task, or domain logic.
The Python distribution is named alethic-kernel; its single import namespace
is alethic.
pip install alethic-kernelfrom alethic import Kernel
kernel = Kernel()
trace = "episode-001"
kernel.write(
"tool",
"percepts",
"COMMIT",
"observation",
{"value": 42, "stale": False, "conflict": False},
trace,
confidence=0.9,
)
proposal = kernel.write(
"planner",
"beliefs",
"PROPOSE",
"threshold_reached",
{"value": True, "depends_on": ["observation"]},
trace,
input_refs=["observation"],
)
committed, reason = kernel.commit_belief_from_proposal(proposal.id, trace)
print(committed, reason)Every worker can propose. Only the kernel can commit. State lives in seven semantic slots: percepts, beliefs, constraints, plans, evidence, predictions, and actions.
This repository contains only the domain-neutral Alethic substrate:
- the governed blackboard kernel and typed record schema;
- evidence, confidence, conflict, constraint, and prediction validation;
- in-memory and SQLite stores;
- worker orchestration, sessions, simulation, and adaptive constraints;
- tests, architectural documentation, and a domain-neutral example.
The controlled study, results, and verification artifacts live separately in governed-cognition.
The repository name is Alethic exactly once. Setuptools maps src/ directly to
the installed alethic namespace.
src/
__init__.py
kernel.py
schema.py
permissions.py
validators.py
store.py
sqlite_store.py
worker.py
orchestrator.py
session.py
sim_worker.py
adaptive_worker.py
python -m pip install -e ".[dev]"
pytest
mypy --strict srcMIT © Emil Uzelac