thinair 0.1.0

Probabilistic Python objects: invent attributes, methods, anything out of thin air. An LLM fills the silence, every answer carries confidence, and written code always wins.

thinair

Probabilistic Python objects. Invent attributes, methods, anything out of thin air; an LLM of your choice (local or hosted) fills in the blanks, with confidence attached.

from thinair import Thing

car = Thing("A Toyota car from the 1990s with a broken engine")

car.color               # "unknown"  (confidence 0.94)
car.year                # 1995       (confidence 0.1 — one year out of a decade)
car.can_drive()         # False      (confidence 0.97)

car.repair_engine()     # no such method — a plan is imagined and *acted out*
car.can_drive()         # True — the object remembers

One axiom: an object is a story, and every interaction is a continuation of it. Everything else falls out — see SPEC.md for the full spec.

The rules

• Written code always wins. Subclass Thing, write real methods and attributes — they run as ordinary Python, byte-for-byte, zero inference. The model is only consulted where your code is silent.
• Certainty is a bare value. Anything from code or explicit assignment is a plain str/int/bool. Inferred values duck-type as their value but carry .confidence — and never claim 1.0.
• Imagined methods can act, not just answer. They read state, write state, and call your real methods (which actually execute). They can never generate or run Python code.

class Boat(Thing):
    """A small motorboat."""
    def refuel(self, litres):
        self.fuel_litres = getattr(self, "fuel_litres", 0) + litres
        return self.fuel_litres

boat = Boat("a dinghy, tank empty")
boat.prepare_for_trip()   # imagined plan → calls the real refuel(20)
boat.fuel_litres          # 20.0 — a bare float, set by real code

Also in the box: stateful=False for memoryless objects (a die that rerolls every read), with Thing.require(0.9): to guard branches on confidence, and freeze()/thaw() to store objects in a database and restore them later.

Setup

No dependencies — one file, stdlib only. Point it at any OpenAI-compatible endpoint (defaults target a local server):

export THINAIR_BASE_URL="http://127.0.0.1:8000/v1"   # default
export THINAIR_API_KEY="1234"
export THINAIR_MODEL="Qwen3.6-35B-A3B-oQ6-mtp"

or in code: Thing.defaults(model="...", base_url="...", api_key="...").

Then:

python demo.py         # the SPEC.md scenes, live
python reddit_bot.py   # an agent that browses reddit through two real methods

Status

An experiment. Every unresolved attribute costs an inference call; answers are as good as your model. That's the fun part.

License

MIT