Ollama Python Library

Warnings

• breaking pip install ollama installs only the Python client — not the Ollama runtime. If the runtime is not installed and running, all calls raise ConnectionError immediately.
  Fix: Install the Ollama runtime separately from https://ollama.com/download, then run 'ollama serve' (or the desktop app). Verify with: curl http://localhost:11434
• breaking Models must be explicitly pulled before use. Calling chat() with a model that hasn't been pulled raises a ResponseError with status_code 404.
  Fix: Run 'ollama pull <model>' before use, or catch ResponseError and call ollama.pull(model) programmatically.
• breaking Tool calling with stream=True is incomplete. When tools are provided, Ollama buffers the full response and returns it as a single block even with stream=True set — true incremental streaming does not occur.
  Fix: Use stream=False when tools are involved. Track https://github.com/ollama/ollama/issues/9084 for resolution.
• gotcha pip install ollama-python installs a different, unofficial third-party package with a completely different API (ModelManagementAPI, GenerateAPI classes). It is not the official client.
  Fix: Use pip install ollama (no hyphen suffix). Official package is at pypi.org/project/ollama.
• gotcha The Ollama Python library defaults to http://localhost:11434. To connect to a remote Ollama instance, you must pass host explicitly via Client(host='http://<remote>:11434'). The OLLAMA_HOST env var is respected by the runtime but NOT automatically picked up by the Python client.
  Fix: Pass host explicitly: Client(host=os.environ.get('OLLAMA_HOST', 'http://localhost:11434'))
• gotcha The OpenAI-compatible endpoint is at http://localhost:11434/v1 — but this is a separate REST interface, not the native ollama Python library. Using openai SDK against this endpoint requires api_key='ollama' (any non-empty string) as a placeholder.
  Fix: from openai import OpenAI; client = OpenAI(base_url='http://localhost:11434/v1', api_key='ollama')
• gotcha Cloud models (e.g. gpt-oss:120b-cloud, deepseek-v3.1:671b-cloud) require ollama signin and ollama pull before use. They are routed through ollama.com and require an API key when accessed via the cloud API endpoint directly.
  Fix: Run 'ollama signin' first. For direct cloud API access, set host='https://ollama.com' and pass Authorization header with OLLAMA_API_KEY.

Install

• pip install ollama Official Ollama Python client

Imports

• chat

  from ollama import chat
  response = chat(model='gemma3', messages=[{'role': 'user', 'content': 'Hello'}])
  print(response.message.content)

  messages must be a list of dicts with role and content keys. Model name must match an already-pulled model.
• Client (remote/custom host)

  from ollama import Client
  client = Client(host='http://localhost:11434')
  response = client.chat(model='gemma3', messages=[{'role': 'user', 'content': 'Hello'}])

  Use Client() for custom host, headers, or timeouts. Default host is http://localhost:11434.

Quickstart

Basic chat with a locally running model

from ollama import chat

response = chat(
    model='gemma3',
    messages=[{'role': 'user', 'content': 'Why is the sky blue?'}]
)
print(response.message.content)