The Official Python Library

A thin, typed wrapper over the REST API

The ollama package on PyPI (installed with pip install ollama) wraps the same REST endpoints in Python functions: chat, generate, embed, list, pull, show and others. Calling the top-level functions uses a default client pointed at http://localhost:11434, or at whatever OLLAMA_HOST is set to. For a different server, create Client(host="http://gpu-box:11434"); for asyncio code use AsyncClient. Responses are typed objects that also allow dictionary-style access, so both response.message.content and response["message"]["content"] work. Passing stream=True returns an iterator of partial responses. Parameters go in an options dictionary with the same names as Modelfile parameters. Errors from the server, such as a model that is not pulled, raise ollama.ResponseError, whose status_code you can inspect. A JavaScript library with a near-identical shape exists for Node and browsers.

Chat, then stream

Requires a running Ollama server and the model pulled locally.

from ollama import chat, ResponseError

messages = [{"role": "user", "content": "Give one tip for learning recursion."}]

try:
    reply = chat(model="llama3.2", messages=messages, options={"temperature": 0.3})
    print(reply.message.content)
except ResponseError as e:
    print("server error:", e.status_code, e.error)

# Streaming: print pieces as they arrive
for part in chat(model="llama3.2", messages=messages, stream=True):
    print(part.message.content, end="", flush=True)
print()

Keep history in a list you own

Append the assistant's reply to messages after each call (messages.append({"role": "assistant", "content": reply.message.content})) before adding the next user turn. That list is the whole memory of the conversation.

Quick check: How do you make the ollama Python library talk to a server on another machine?

• Pass remote=True to chat()
• Create Client(host="http://other-machine:11434") and call its methods
• Edit the model's Modelfile
• It is impossible; the library only works locally