Together Python Client

2.7.0 · active · verified Thu Apr 09

The Together Python library (v2.x) provides convenient access to the Together AI REST API for Python 3.9+ applications. It offers strongly-typed request parameters and response fields, with both synchronous and asynchronous clients powered by httpx. This modern SDK is generated from the OpenAPI specification using Stainless, ensuring a 1:1 mapping to the API and rapid feature delivery. Version 1.x is now in maintenance mode, with all new development focused on v2.x.

Warnings

breaking The Together Python SDK v2.0 (released December 2025) introduced significant breaking changes from v1.x. Key areas affected include constructor parameters (e.g., `supplied_headers` changed to `default_headers`), error handling class internals and exceptions, mandatory keyword arguments for all API calls (no positional parameters), and changes to many response type names in `together.types.*`.
Fix: Refer to the official Python SDK Migration Guide for detailed API-by-API notes and code snippets. Update all API calls to use keyword arguments and adjust error handling and type imports accordingly.
gotcha API calls in v2.x and later strictly require all parameters to be passed as keyword arguments. Positional arguments are no longer supported. Attempting to use positional arguments will result in runtime errors.
Fix: Always use `param=value` syntax for all arguments when calling methods such as `client.chat.completions.create()`.
gotcha For streaming responses (e.g., chat completions), the default `create` method eagerly reads the full response. To stream tokens incrementally as they are generated by the model, you must explicitly use the `.with_streaming_response` method within a context manager.
Fix: Wrap streaming calls with `with client.chat.completions.with_streaming_response(...) as stream_response:` and then iterate over `stream_response.iter_bytes()`, `iter_text()`, or `iter_lines()` for efficient token handling.
gotcha The library extensively uses Pydantic for type validation and data modeling. If your project integrates `together-py` with other Python libraries (e.g., LangChain, FastAPI) that have differing or strict Pydantic v1 or v2 dependencies, you might encounter compatibility issues. Pydantic v2 is a major rewrite and not fully backward compatible with v1.
Fix: Ensure all libraries in your project are compatible with a single Pydantic version (preferably v2). If conflicts are unavoidable, consider using tools like `pydantic-compat` or isolating environments, though direct compatibility is ideal. `together-py` v2 is designed for Pydantic v2 environments.

Install

pip install together Install via pip
uv add together Install via uv (recommended)

Imports

Together
```
from together import Together
```
AsyncTogether
```
from together import AsyncTogether
```

Quickstart

Demonstrates how to initialize the synchronous client and make a basic chat completion request using an environment variable for the API key.

import os
from together import Together

# Ensure TOGETHER_API_KEY is set in your environment variables
# e.g., export TOGETHER_API_KEY="your_api_key_here"
client = Together(
    api_key=os.environ.get("TOGETHER_API_KEY", "")
)

try:
    chat_completion = client.chat.completions.create(
        messages=[
            {"role": "user", "content": "Say this is a test!"}
        ],
        model="meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo",
    )
    print(chat_completion.choices[0].message.content)
except Exception as e:
    print(f"An error occurred: {e}")
    print("Please ensure TOGETHER_API_KEY is set and valid, and you have access to the specified model.")

view raw JSON →