logical

A triple-backed logic engine for turning natural language knowledge into RDF-like claims, projecting accepted claims into Prolog, and asking logical questions over that generated world.

First developed at the OpenAI emergency hackathon on 3/5/2023.

Install

brew install swi-prolog
uv sync --dev
cp .env-example .env

Set OPENAI_API_KEY in .env for commands that call OpenAI. The default model is gpt-5.5.

Usage

uv run logical add "the sky is red"
uv run logical add "a sky can only be one color"
uv run logical add "the sky is blue"
uv run logical ask "is the sky red?"
uv run logical check
uv run logical export-prolog

logical add and logical ask require OPENAI_API_KEY because they use OpenAI to extract structured triples or query intent. logical check and logical export-prolog operate on the local .logical/ store and do not need an API key.

Knowledge is stored in .logical/knowledge.jsonl. Prolog is generated into .logical/world.pl; it is a projection, not the source of truth.

Invalid or unsupported logic is quarantined instead of projected into Prolog. Conflicting claims are also quarantined by default. In an interactive terminal, logical add --interactive "..." lets you choose whether to keep existing knowledge, replace it, or quarantine the new claim.

Development

uv run pytest

The tests use fake extractors, so they do not require OPENAI_API_KEY. The suite includes a deterministic 100-example coverage pass for valid knowledge, invalid logic, conflicts, conflict resolution, Prolog projection, and query behavior. Tests that require SWI-Prolog skip with a clear message when swipl is not on PATH.