Ollama Python SDK
https://github.com/ollama/ollama-python- Category
- Developer Tools
- Rank
- No. 178Tools index
- Pricing
- Open Source
- Type
- TOOL
- Builder
- ollama
- GitHub
- 10.5k stars
- Latest release
- v0.6.2
- Date
About
Official Python library for Ollama — run, manage, and chat with locally-served LLMs from Python.
What it does
This package turns Ollama’s REST endpoints into typed Python objects and straightforward synchronous or asynchronous requests. It handles conversations, text generation, embeddings, model administration, streaming output, tool descriptions, multimodal image input, web search, and web fetching. Clients can target a local service or Ollama’s cloud API.
Why it's ranked here
This is a strong default for Python projects already committed to Ollama. Its API covers inference and model lifecycle work, supports streaming in both synchronous and asynchronous code, and validates request and response data with Pydantic. The narrow dependency set and direct REST mapping keep the abstraction understandable.
What's good
Responses support both attribute and mapping-style access, easing typed adoption without forcing one access pattern. Streaming yields validated response objects and raises explicit errors during the stream. Custom clients accept transport settings such as headers, host, redirects, and timeouts. Python functions can also be converted into model tool schemas from signatures and docstrings.
Tradeoffs
The package is a client, so local use still requires Ollama installed, running, and supplied with a pulled model. Direct cloud access requires an API key, while cloud use through local Ollama requires sign-in and pulling a cloud model. Tool argument values may not exactly match their declared schema, and the example explicitly casts them before execution.
How to use it well
Use it in Python services, scripts, notebooks, or asynchronous applications that need typed access to an existing Ollama endpoint. Choose streaming for incremental output and a custom client for remote hosts or authentication. Keep server installation, model selection, model acquisition, and tool execution logic outside the library.
Technical notes+
pyproject.toml declares Python >=3.8, MIT licensing, httpx>=0.27, pydantic>=2.9, and a Hatchling build with VCS-derived versions. ollama/_client.py implements Client and AsyncClient over httpx, maps methods to /api endpoints, parses streamed JSON lines, and wraps connection or HTTP failures. ollama/_types.py defines Pydantic request and response models, mapping-style model access, image serialization, tool schemas, thinking controls, log probabilities, and generation fields. ollama/_utils.py builds Tool schemas from callable signatures and Google-style docstrings. ollama/__init__.py exports both client classes, response types, errors, and module-level methods backed by a shared Client instance. requirements.txt provides a hash-pinned exported environment.
Observed
- License
- MIT
- Primary language
- Python
- Python support
- Python 3.8 or newer
- Installation
- Published package installed with pip install ollama
- Interface
- Python library client for the Ollama REST API, with synchronous and asynchronous clients
- Runtime dependencies
- httpx and Pydantic
- Build system
- Hatchling with VCS-derived package versions
Read from README.md, pyproject.toml, requirements.txt, ollama/_types.py, ollama/_utils.py, ollama/_client.py, ollama/__init__.py, examples/ps.py, examples/chat.py, examples/list.py, examples/pull.py, examples/show.py, examples/embed.py, examples/tools.py, examples/create.py.
What it can do
Run local LLM inference
Text prompt and model name → Generated text response
Chat with LLM in conversation format
Conversation history and new message → Contextual chat response
List available local models
API request → Array of installed model names and details
Pull and download LLM models
Model name/identifier → Downloaded model ready for use
Delete local LLM models
Model name → Model removed from local storage
Stream text generation responses
Text prompt with streaming enabled → Real-time text tokens as they generate
Generate embeddings from text
Text string and embedding model → Numerical vector representation
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.