clipspyx 0.7.1

CLIPS Python bindings via CFFI

clipspyx

Python bindings and DSL for CLIPS, the production rule engine. Write rules as Python classes, run them with an async goal handler loop, trace fact derivation chains, and visualize your rule network as D2 diagrams.

Originally forked from clipspy, clipspyx adds a Python-native DSL, async backward chaining, fact tracing, and full CLIPS 7.0 support while keeping the low-level CFFI API available for everything else.

Quick start

pip install clipspyx          # or: uv add clipspyx

Low-level API

from clipspyx import Environment

env = Environment()
env.build('(defrule hello (initial-fact) => (println "Hello CLIPS!"))')
env.reset()
env.run()

Python DSL

Define CLIPS templates and rules as Python classes:

from clipspyx import Environment
from clipspyx.dsl import Template, Rule, Multi

class Employee(Template):
    name: str
    title: str
    years: int = 0

class Skill(Template):
    name: str
    proficiency: int = 1

class Project(Template):
    name: str
    required_skill: str
    min_proficiency: int = 1

class AssignProject(Rule):
    """Match employees to projects by skill."""
    e = Employee(name=name)
    Skill(name=skill, proficiency=prof)
    Project(name=proj, required_skill=skill, min_proficiency=min_prof)
    prof >= min_prof

    def __action__(self):
        print(f"Assign {self.name} to project {self.proj}")

env = Environment()
NewEmployee = env.define(Employee)
NewSkill = env.define(Skill)
NewProject = env.define(Project)
env.define(AssignProject)

env.reset()
NewEmployee(name="Alice", title="Engineer", years=5)
NewSkill(name="Python", proficiency=4)
NewProject(name="Atlas", required_skill="Python", min_proficiency=3)
env.run()
# => Assign Alice to project Atlas

The DSL supports the full range of CLIPS conditional elements:

Python syntax	CLIPS CE
p = Person(name=name)	Assigned pattern
Person(age=_)	Wildcard
Person(age=25)	Literal constraint
Person(name=not "Bob")	Field negation
Person(age=25 or 30)	Field or
age >= 18	Test CE
~Person(name="Bob")	Negation
exists(Person())	Exists CE
forall(Person(name=n), Badge(owner=n))	Forall CE
logical(Person(name=n))	Logical CE

Rules can declare effects without a Python bridge function:

class DoubleInput(Rule):
    i = Input(value=v)
    asserts(Output(value=v * 2))

class Cleanup(Rule):
    p = Temp(flag=Symbol("done"))
    retracts(p)

Fact tracing

Track which rules produced which facts:

env.enable_tracing()
env.reset()
# ... assert facts and run ...

for firing in env.find_template('RuleFiring').facts():
    print(f"{firing['rule']}: {firing['inputs']} -> {firing['outputs']}")

Each RuleFiring fact records the rule name, the input facts that matched, and the output facts that were asserted.

Visualization

Generate D2 diagrams of your rule network:

# Print D2 text
print(env.visualize(group_by_kind=True))

# Render to SVG (requires d2 CLI)
env.visualize(output="diagram.svg", group_by_kind=True)

Templates, rules, pattern edges, fact-address references, and effect edges are all represented. See examples/hr_system.py for a full example.

Async goal handlers (CLIPS 7.0)

CLIPS 7.0 backward chaining generates goals when a rule needs a fact that doesn't exist yet. clipspyx dispatches these goals to Python async handlers, enabling rules that wait on timers, HTTP calls, or any async I/O:

import asyncio
from clipspyx import Environment
from clipspyx.dsl import Template, Rule, TimerEvent, AFTER
from clipspyx.values import Symbol

class Alert(Template):
    msg: str

class DelayedAlert(Rule):
    te = TimerEvent(kind=AFTER, name=Symbol("alarm"), seconds=2.0)
    asserts(Alert(msg=Symbol("time-is-up")))

env = Environment()
env.enable_goal_handlers()
env.define(Alert)
env.define(DelayedAlert)
env.reset()
asyncio.run(env.async_run())

Built-in TimerEvent supports one-shot (AFTER), absolute (AT), and periodic (EVERY) timers. Register custom handlers for your own goal templates:

async def query_handler(goal, env):
    result = await fetch_from_service(goal['service'], goal['query'])
    env.find_template('QueryResult').assert_fact(result=result)

env.register_goal_handler('query-goal', query_handler)

CLIPS 7.0 features

When built against CLIPS 7.0x, clipspyx also supports:

• Template inheritance: (deftemplate car (is-a vehicle) (slot doors))
• Certainty factors: (deftemplate sensor (is-a CFD) (slot reading)) with CF propagation through rule chains
• Deftables: static lookup tables accessible from rules and Python
• Backward chaining: goal generation with (goal ...) patterns
• update_slots: selective rete re-evaluation (only rules matching the changed slot)

Building from source

clipspyx compiles CLIPS source directly into the CFFI extension. No separate libclips build step.

Prerequisites

• Python 3.10+
• C compiler (gcc/clang on Linux/macOS, MSVC on Windows)
• CLIPS source (auto-checked out from orphan branch, or provide your own)

Install for development

# Install with 64x backend (CLIPS 6.4x)
uv sync --extra 64x

# Install with 70x backend (CLIPS 7.0x)
uv sync --extra 70x

Switching backends

The 64x and 70x backends conflict with each other. Always use uv sync to switch:

uv sync --extra 64x   # switch to 6.4x
uv sync --extra 70x   # switch to 7.0x

uv run --extra does not remove the previously installed conflicting backend. Only uv sync --extra <variant> correctly handles the switch.

CLIPS source override

CLIPS_SOURCE_DIR=/path/to/clips/core uv sync --extra 64x

Build distributable wheels

uv run scripts/build-backend.py 64x   # builds clipspyx-ffi + clipspyx-ffi-64x wheels
uv run scripts/build-backend.py 70x   # builds clipspyx-ffi + clipspyx-ffi-70x wheels

Testing

uv run pytest tests/ -v

License

BSD-3-Clause. See LICENSE.txt.