Determine

Determine makes AI inference feel like a normal part of Python.

The main API has only two operations:

determine.answer(...)
determine.choose(...)

Install

pip install determine

Ollama

Run an installed Ollama model directly:

import determine

determine.configure("ollama:qwen3")

name = "Jack"

print(
    determine.answer(
        "Say hello to the user."
    )
)

If you leave the model out:

determine.configure("ollama")

Determine will try to use the first installed Ollama model.

llama.cpp

Use a local GGUF directly:

import determine

determine.configure(
    "llama.cpp:/home/me/models/model.gguf"
)

print(
    determine.answer(
        "What should I do?"
    )
)

Determine automatically looks for either:

llama-cli

or:

llama cli

OpenAI-compatible APIs

Determine still supports OpenAI-compatible endpoints:

import determine

determine.configure(
    "http://localhost:8080"
)

answer()

Use answer when the result can be anything:

answer = determine.answer(
    "What should the player be told?"
)

choose()

Use choose when the result must come from available options:

action = determine.choose(
    "What should the enemy do?",
    [
        "attack",
        "run",
        "hide"
    ]
)

Automatic context

Determine can automatically see relevant variables and source code.

For example:

health = 12
ammo = 0
enemy_distance = 3

action = determine.choose(
    "What should the player do?",
    [
        "fight",
        "run",
        "hide"
    ]
)

Those variables do not have to be manually placed into the prompt.

Advanced llama.cpp configuration

Extra llama.cpp arguments can be supplied:

determine.configure(
    "llama.cpp:/home/me/model.gguf",
    temperature=0.2,
    reasoning_effort="high",
    args=[
        "-ngl", "all",
        "-c", "32768"
    ]
)

A custom executable can also be used:

determine.configure(
    "llama.cpp:/home/me/model.gguf",
    binary="/home/me/llama.cpp/build/bin/llama-cli"
)

Functions as choices

Functions can be choices too:

def backup():
    """Back up the phone."""
    print("Backing up")

def update(version="latest"):
    """Update the phone."""
    print("Updating to", version)

request = input("> ")

determine.choose(
    request,
    [
        backup,
        update
    ]
)

Security

Determine can make fuzzy decisions, but important security boundaries and destructive actions should still have deterministic validation and confirmation around them.

License

MIT

Multi-turn conversations

Determine can remember previous turns by giving it a normal Python list.

import determine

determine.configure("ollama:qwen3")

history = []

print(
    determine.answer(
        "My favourite colour is green.",
        history=history
    )
)

print(
    determine.answer(
        "What is my favourite colour?",
        history=history
    )
)

The same history can be shared with choose():

history = []

determine.answer(
    "I prefer quiet places.",
    history=history
)

place = determine.choose(
    "Where should I go?",
    [
        "busy shopping centre",
        "quiet park",
        "concert"
    ],
    history=history
)

print(place)

Without history=, Determine remains single-turn.

You can start a fresh conversation simply by creating a new list:

history = []