clean-ollama 1.2.2

A minimal, typed Python interface for Ollama that simplifies the existing ollama library

clean-ollama

A minimal, typed Python interface for Ollama that simplifies the existing ollama library

Installation

Install with pip

pip install clean-ollama

Requirements

• Python 3.10+
• Ollama installed

Usage

Basic generation

from clean_ollama import Client, Message, Role

client = Client("qwen3.5:4b")
messages = [
    Message(Role.USER, "What is the capital of France?")
]
thinking, response, tool_calls = client.generate(messages)
print(response)

Streaming

from clean_ollama import Client, Message, Role

client = Client("qwen3.5:4b")
messages = [
    Message(Role.USER, "Tell me a joke.")
]
for thinking, chunk in client.stream(messages):
    print(chunk, end="", flush=True)

Tool use

from clean_ollama import Client, Message, Role, Tool, Param, ParamType

get_weather = Tool(
    name="get_weather",
    description="Get the current weather for a location",
    params=[
        Param("location", "City name", ParamType.string),
        Param("units", "celsius or fahrenheit", ParamType.string, required=False),
    ]
)
messages = [
    Message(Role.USER, "What's the weather in Tokyo?")
]
thinking, response, tool_calls = client.generate(messages, tools=[get_weather])
print(tool_calls)

What tool calls look like

generate returns a list of tool calls. Each call has a function with a name and arguments dict:

thinking, response, tool_calls = client.generate(messages, tools=[get_weather])
for call in tool_calls:
    print(call.function.name) # "get_weather"
    print(call.function.arguments) # {"location": "Tokyo", "units": "celsius"}

If the model didn't call any tools, tool_calls is an empty list.

Thinking (extended reasoning)

from clean_ollama import Client, Message, Role

thinking, response, tool_calls = client.generate(messages, think=True)
print("Thinking:", thinking)
print("Response:", response)

Loading and unloading models

from clean_ollama import Client

client = Client("qwen3.5:4b")
client.load() # keeps the model in memory
client.unload() # frees the model from memory

API Reference

Client(model: str)

Main class for interacting with Ollama.

Method	Description
load()	Loads the model into memory
unload()	Unloads the model from memory
generate(messages, tools=None, think=False)	Returns (thinking, response, tool_calls)
stream(messages, think=False)	Yields (thinking, chunk) pairs

Message(role, content)

Represents a chat message. role accepts a Role enum or a plain string.

Role

Enum with values: SYSTEM, USER, ASSISTANT, TOOL

Tool(name, description, params)

Defines a callable tool the model can invoke.

Param(name, description, param_type, required=True)

Defines a single tool parameter.

ParamType

Enum for parameter types: string, integer, float, boolean, json

License

MIT