LlamaIndex

Install and connect

pip install llama-index-core llama-index-llms-openai-like
export NOKEEP_API_KEY="sk-nokeep-test"  # your own key once it connects

import os
from llama_index.llms.openai_like import OpenAILike

llm = OpenAILike(
    model="deepseek-v4.1-flash-uncensored",
    api_base="https://nokeep.ai/v1",
    api_key=os.environ["NOKEEP_API_KEY"],
    is_chat_model=True,
    is_function_calling_model=True,
    context_window=1000000,
    temperature=0.8,
)

print(llm.complete("Name three good names for a tavern.").text)

• is_chat_model=True sends chat completions. Without it, LlamaIndex calls /v1/completions, which nokeep doesn’t have.
• is_function_calling_model=True lets agents and chat_with_tools send tools.
• context_window defaults to 3,900 tokens, so LlamaIndex trims long chats and retrieved text to fit. Set it to the model’s real context: 1,000,000 for DeepSeek, 128,000 for GLM.
• temperature defaults to 0.1, which makes fiction flat and repetitive. Set your own.

Chat and stream

from llama_index.core.llms import ChatMessage

messages = [
    ChatMessage(role="system", content="You are a blunt editor."),
    ChatMessage(role="user", content="Cut this opening line to ten words: ..."),
]
print(llm.chat(messages).message.content)

for part in llm.stream_chat(messages):
    print(part.delta, end="", flush=True)

Tools and agents

import asyncio
from llama_index.core.agent.workflow import FunctionAgent

def get_weather(city: str) -> str:
    """Get the weather in a city."""
    return f"18°C and clear in {city}"

agent = FunctionAgent(
    tools=[get_weather],
    llm=llm,
    system_prompt="You answer questions about the weather.",
)

async def main():
    print(await agent.run("What's the weather in Lisbon?"))

asyncio.run(main())

For one call with tools and no loop, use llm.chat_with_tools([tool], user_msg=...) and read the calls with llm.get_tool_calls_from_response(...). See Tool calls.

Thinking and other fields

import os
from llama_index.llms.openai_like import OpenAILike

thinker = OpenAILike(
    model="deepseek-v4.1-flash-uncensored",
    api_base="https://nokeep.ai/v1",
    api_key=os.environ["NOKEEP_API_KEY"],
    is_chat_model=True,
    context_window=1000000,
    max_tokens=4000,
    additional_kwargs={"reasoning_effort": "high", "min_p": 0.05},
)

additional_kwargs goes into every request as it is. Thinking counts toward max_tokens, so give it room. See Thinking.

Gotchas

• `reasoning_effort` takes only the model’s own values. deepseek-v4.1-flash-uncensored takes none, low, high, max; glm-5.3-flash-uncensored takes low, high. Other apps’ minimal, medium and xhigh run at the model’s nearest level (medium as high, xhigh as max where the model has it); any other value gets 400 invalid_reasoning_effort. See Thinking.
• GLM picks its own tools. glm-5.3-flash-uncensored takes tool_choice auto or none only. Forcing a tool (required, or naming a function) gets 400 unsupported_tool_choice. Use DeepSeek when you need to force one.
• Fields a model can’t take are dropped, not refused. The call still answers, and the x-nokeep-ignored response header names what was left out. GLM keeps only tools, tool_choice and reasoning_effort, so its temperature and response_format are dropped. See which fields each model takes.
• `n` must be 1. Ask for several replies with several calls.
• Chat completions only. There is no /v1/responses and no /v1/embeddings; a framework that defaults to either needs to be pointed at chat completions, and embeddings need another provider.

For a RAG app, set Settings.llm = llm and keep your embedding model where it is: nokeep runs the chat model, not the embeddings. With glm-5.3-flash-uncensored, the temperature LlamaIndex sends on every call is dropped and named in x-nokeep-ignored.