How do I install the Ollama Python library?

Run pip install ollama with Python 3.8 or newer. Then import ollama and call ollama.chat(...) . A local Ollama server must be running for the calls to succeed.

How do I stream responses with the Python client?

Pass stream=True to ollama.chat . It returns a generator; iterate it and print chunk['message']['content'] with end='' so tokens appear as they arrive instead of all at once.

What is the difference between chat and generate?

ollama.chat takes a messages list with roles for multi-turn conversation. ollama.generate takes a single prompt for one-off completions. Chat streams expose message.content ; generate streams expose response .

How do I connect to a remote Ollama server in Python?

Create a client with a host: Client(host="http://box:11434") , then call .chat(...) on it. Use AsyncClient the same way inside async code for concurrent requests.

Build / AI / Ollama

The Ollama Python Library

Use the official ollama Python library to chat, stream tokens, manage models, and call a remote host.

TL;DR

Install the official client with pip install ollama.
Call ollama.chat() with a messages list to get replies.
Stream tokens as they arrive by passing stream=True.

Jump to section

Install And Chat
Streaming
Manage Models
Clients
Tips
Warnings
In Practice
FAQ

Install And Chat

Install

Add the official Python client.

pip install ollama

chat

Send messages and read the reply.

import ollama
r = ollama.chat(model="llama3.2",
  messages=[{"role":"user","content":"Hi"}])

Read Content

The reply text is on message.content.

print(r["message"]["content"])

generate

Send a single prompt without roles.

ollama.generate(model="llama3.2",
  prompt="Explain DNS")

Streaming

stream=True

Return a generator of partial chunks.

stream = ollama.chat(model="llama3.2",
  messages=msgs, stream=True)

Iterate Chunks

Print each token as it arrives.

for c in stream:
  print(c["message"]["content"], end="")

generate Stream

Chunks expose response for generate.

# c["response"] for generate streams

Manage Models

list

List locally installed models.

ollama.list()

pull

Download a model from Python.

ollama.pull("llama3.2")

embed

Create embeddings from text.

ollama.embed(model="nomic-embed-text",
  input="hello")

show

Inspect a model's details.

ollama.show("llama3.2")

Clients

Client

Point the client at a specific host.

from ollama import Client
c = Client(host="http://box:11434")

AsyncClient

Use async for concurrent requests.

from ollama import AsyncClient
r = await AsyncClient().chat(...)

tools

Pass Python functions as tools.

ollama.chat(model="llama3.1",
  messages=msgs, tools=[get_weather])

Tips

Iterate the generator from stream=True and print chunk['message']['content'], so output appears token by token like a live chat.
Use AsyncClient for concurrent requests in async apps, or Client(host=...) to target a remote Ollama server by URL.

Warnings

The client needs a running server; start ollama serve or the desktop app, or calls fail to reach localhost:11434.
ollama.chat is stateless like the API; resend the full messages list on each call to keep conversation context.

In Practice

Send a prompt with the Python client and print the reply token by token as it streams in.

pip install ollama gives the official client.
stream=True makes chat return a generator of chunks.
Each chunk holds a slice of text on message.content.
Printing with end="" streams the reply smoothly.

import ollama

stream = ollama.chat(
    model="llama3.2",
    messages=[{"role": "user",
               "content": "Explain RAG in 3 lines"}],
    stream=True,
)

for chunk in stream:
    print(chunk["message"]["content"],
          end="", flush=True)