langchain-linkrra

LangChain chat model for Linkrra Synap — an OpenAI-compatible inference API for open models (Qwen, gpt-oss, Gemma, GLM, DeepSeek and more), billed per token.

Not to be confused with other products named "Synap". This package talks to https://synap402.com/v1, operated by Linkrra.

Install

pip install -U langchain-linkrra
export LINKRRA_API_KEY="your-api-key"   # get one at https://synap402.com

Use

from langchain_linkrra import ChatLinkrra

model = ChatLinkrra(model="gpt-oss-20b", temperature=0)
print(model.invoke("Write a haiku about open models.").content)

model="synap-v1" (the default) lets Synap pick the cheapest model that can do the job. The live model list, with context lengths and per-token prices, is always at https://synap402.com/v1/models — prices are not hard-coded here.

ChatLinkrra subclasses LangChain's BaseChatOpenAI, so streaming, async, bind_tools(...) and with_structured_output(...) work as with ChatOpenAI.

from pydantic import BaseModel

class City(BaseModel):
    name: str
    country: str

structured = ChatLinkrra(model="qwen3-coder-30b").with_structured_output(City)
structured.invoke("Which city is the Eiffel Tower in?")

Configuration

Tests

pip install -e ".[test]"
pytest tests/unit_tests                 # offline
LINKRRA_API_KEY=... pytest tests/integration_tests   # one real call

License

Apache-2.0