langchain-obscurify

LangChain chat model for Obscurify.ai, a private AI chat service with an OpenAI-compatible API (200+ models).

ChatObscurify supports invoke, streaming, async, tool calling, structured output and reasoning effort.

Install

pip install -U langchain-obscurify

Get an API key

1. Create an account at obscurify.ai. An account is a random account number: no email, username or phone.
2. Add a balance. The API works on every plan with a positive balance. New accounts get $1 of starter credit (limits apply; see https://obscurify.ai/pricing).
3. Open https://obscurify.ai/account and click Show API Key. Keys look like sk-obs1_....

export OBSCURIFY_API_KEY="sk-obs1_..."

Usage

from langchain_obscurify import ChatObscurify

model = ChatObscurify(model="deepseek/deepseek-v4-flash", temperature=0)

print(model.invoke("Write a haiku about privacy.").content)

for chunk in model.stream("Count to five."):
    print(chunk.text, end="", flush=True)

Tool calling

from langchain_core.tools import tool


@tool
def get_weather(city: str) -> str:
    """Get the current weather for a city."""
    return f"It is sunny in {city}."


msg = model.bind_tools([get_weather]).invoke("What's the weather in Paris?")
print(msg.tool_calls)

Structured output

from pydantic import BaseModel


class Joke(BaseModel):
    setup: str
    punchline: str


print(model.with_structured_output(Joke).invoke("Tell me a joke about cats"))

Reasoning

model = ChatObscurify(model="z-ai/glm-5.3-flash", reasoning_effort="low")
msg = model.invoke("Is 221 prime?")
print(msg.additional_kwargs.get("reasoning_content"))
print(msg.content)

Agents

from langchain.agents import create_agent

agent = create_agent(ChatObscurify(model="deepseek/deepseek-v4-flash"), tools=[get_weather])
agent.invoke({"messages": [{"role": "user", "content": "Weather in Oslo?"}]})

Models

Any model ID from GET https://obscurify.ai/v1/models works, for example deepseek/deepseek-v4-flash, meta-llama/llama-3.3-70b-instruct, z-ai/glm-5.3-flash or qwen/qwen3.8-27b. Frontier models need the Casual plan; uncensored models need Pro. Prices per model: https://obscurify.ai/model_cards.

Configuration

All other ChatOpenAI arguments (temperature, max_tokens, timeout, max_retries, reasoning_effort, model_kwargs, ...) work the same way.

Privacy

Obscurify.ai doesn't store prompts or responses, scrambles identities before requests reach model providers, and sends text requests only over zero-data-retention routes. Details: https://obscurify.ai/privacy.

Links

• API docs: https://obscurify.ai/api-docs
• Source: https://github.com/Obscurify-ai/langchain-obscurify
• JavaScript package: @obscurify/langchain
• Contact: https://obscurify.ai/contact

MIT licensed.