PyAgentProlog

A Prolog engine in Python with proof tracing, agents based on Python actions, an expert system, and semantic unification on embeddings.

Author

Andrey Vorfolomeev

Features

• Prolog core: parser, iterative solver with TCO (tail-call optimization), cut (!), if-then-else, rule indexing by the first argument.
• Proof tracing — an inference tree with unifications and rules.
• Numba-accelerated arithmetic — a non-recursive stack machine, runs on ARM/Raspberry Pi.
• Agents — Python functions as built-in predicates (@simple_action, @effector_action, @generator_action).
• Expert system — interactive mode, extraction of conditions from natural language via pymorphy3, contradiction resolution, glossary.
• Semantic unification — matching atoms via embeddings (Ollama or sentence-transformers).
• Integration with LangChain — PrologChain as Runnable in LCEL chains.
• Gradio sessions — for interactive UIs.

Installation

Basic (core only)

pip install pyagentprolog

With additional features

# Numba acceleration for arithmetic
pip install pyagentprolog[numba]

# Semantic unification
pip install pyagentprolog[embeddings]

# LangChain
pip install pyagentprolog[langchain]

# Expert system (pymorphy2)
pip install pyagentprolog[expert]

# Everything at once
pip install pyagentprolog[all]

Quick start

Simple query

from prolog import Solver

program = """
parent(ivan, alexey).
parent(alexey, petr).

grandfather(D, V) :-
    parent(D, R),
    parent(R, V).
"""

solver = Solver(program)
solutions = solver.find_solutions("grandfather(ivan, X).")
print(solutions)  # [{'X': 'petr'}]

Proof trace

solutions, explanation = solver.find_solutions("grandfather(ivan, X).",
                                                with_explanation=True)
print(solver.get_explanation())

Agent with Python actions

from prolog import Agent, simple_action

@simple_action('add', 3, output_arg=2)
def add(a, b, context=None):
    return a + b

agent = Agent(program="run(X) :- add(2, 3, X).")
agent.register_action(add)
print(agent.query_one("run(X)."))  # {'X': 5}

### Expert System

Requires pyagentprolog[expert]

```python
from prolog import Expert

expert = Expert(domain_file="knowledge.pl", use_pymorphy=True)
answer = expert.ask("I want to learn programming")
print(answer)

Requirements

• Python 3.9+
• Mandatory: numpy
• Optional: numba, langchain-core, sentence-transformers, langchain-ollama, gradio, pymorphy2

License

MIT — see the LICENSE file.