InferLingo

InferLingo is a small Python rule engine for facts and rules written as readable sentences. Use deterministic code to establish facts, InferLingo to derive consequences from explicit rules, and optional semantic unification when differently worded sentences may express the same fact.

InferLingo is not a chatbot, planner, arithmetic engine, source-code analyzer, or action runner. Python or another deterministic subsystem supplies facts. InferLingo applies the rules you write, builds inspectable proofs, and returns derived consequences.

Install

Install the base package for deterministic inference:

python -m pip install inferlingo

The base package is fully usable offline. The Python KnowledgeBase API defaults to ExactUnifier, and the first CLI workflows should use --exact-only.

Install the optional semantic backend only when differently worded sentences should be compared:

python -m pip install 'inferlingo[jev]'

Authentication and provider configuration belong to pyjev and its configured Jev backend.

60-second exact example

Create rules.nl:

Alice is an employee.
Bob is an employee.
Bob is suspended.

{person} may deploy if
    {person} is an employee and
    not {person} is suspended.

Run it without a model or network access:

inferlingo run rules.nl "{person} may deploy?" --exact-only --explain

The solution binds person = Alice. The result follows only from the facts and explicit rule. Negation is failure to prove the positive goal for the already-bound person, not a stored classical negative fact.

The smallest chain is also available:

inferlingo run examples/birds.nl "{bird} can fly?" --exact-only --explain

It derives bird = Tweety from the fact and rules in examples/birds.nl, not from outside knowledge.

How it fits into Python

from inferlingo import KnowledgeBase

kb = KnowledgeBase.from_text(
    """
    Alice is an employee.
    {person} may deploy if {person} is an employee.
    """
)
result = kb.ask_sync("{person} may deploy?")
print(result.solutions[0].bindings)

The result is {"person": "Alice"}. Use await kb.ask(...) in asynchronous applications. ask_sync() must not be called from a running event loop.

KnowledgeBase.add_fact() safely quotes injected application values and can carry Provenance metadata into proof steps:

from inferlingo import KnowledgeBase, Provenance

kb = KnowledgeBase.from_text("URL {url} is approved if URL {url} is reachable.")
kb.add_fact(
    "URL {url} is reachable",
    url="https://example.com/a?x=1&y=2",
    provenance=Provenance(source="scanner.py", line=41, kind="http-check"),
)

Optional semantic unification

PyJevUnifier first tries exact wording and only sends differently worded candidates to the semantic backend. Semantic matching is a same-fact equivalence check. It does not invent logical implications. An explicit rule is still required to derive parent from father.

python -m pip install 'inferlingo[jev]'
pyjev auth set
pyjev auth test
inferlingo run examples/family.nl "Homer is a parent of Lisa?" --explain

The CLI uses PyJevUnifier by default for run; pass --exact-only to force deterministic offline execution. Semantic confidence and backend checks are diagnostic signals, not theorem probabilities.

Documentation

Read the full documentation for:

• Getting started
• Language reference
• Python API
• CLI guide
• Semantic unification
• Proofs and provenance
• Examples

The documentation also covers strict safety validation, quoted atoms, recursion, search limits, traces, debugging, development, and the generated API reference.

Development

For a source checkout, install contributor and documentation dependencies:

python -m pip install -e '.[dev,docs]'

Run the quality checks:

python -m compileall inferlingo examples docs
pytest
ruff check .
python docs/make.py html
python -m build
twine check dist/*

The test suite is offline. Semantic tests use fake or injected clients and do not require a live API key.

License

Apache-2.0