Feature request: native async client (AsyncZaiClient)

[clemlesne]

Problem

The zai-sdk currently only provides synchronous clients (ZaiClient, ZhipuAiClient). When used in an async Python application (e.g. FastAPI, aiohttp), every SDK call blocks the event loop thread.

Workarounds required today

Non-streaming calls must be wrapped in asyncio.to_thread():

response = await asyncio.to_thread(
    client.chat.completions.create,
    model="glm-4",
    messages=messages,
)

Streaming calls are worse — not only the initial request but also each __next__() call on the stream iterator blocks. This requires a background thread + asyncio.Queue bridge:

stream = await asyncio.to_thread(client.chat.completions.create, ..., stream=True)

queue = asyncio.Queue()
loop = asyncio.get_running_loop()

def _iter_in_thread():
    try:
        for chunk in stream:
            loop.call_soon_threadsafe(queue.put_nowait, chunk)
    except Exception as exc:
        loop.call_soon_threadsafe(queue.put_nowait, exc)
    finally:
        loop.call_soon_threadsafe(queue.put_nowait, None)

loop.run_in_executor(None, _iter_in_thread)

while True:
    item = await queue.get()
    if item is None:
        break
    if isinstance(item, Exception):
        raise item
    yield item

This adds ~30 lines of thread-safety boilerplate per provider, increases thread pool pressure, and makes error propagation fragile.

How other SDKs solve this

All major LLM SDKs provide a native async client alongside the sync one:

SDK	Sync client	Async client
openai	OpenAI	AsyncOpenAI
anthropic	Anthropic	AsyncAnthropic
groq	Groq	AsyncGroq
mistralai	Mistral	AsyncMistral (via httpx.AsyncClient)
cerebras-cloud-sdk	Cerebras	AsyncCerebras
zai-sdk	ZaiClient / ZhipuAiClient	❌ missing

Proposal

Add AsyncZaiClient and AsyncZhipuAiClient with the same API surface but using httpx.AsyncClient under the hood.

from zai import AsyncZaiClient

client = AsyncZaiClient(api_key="...")

# Non-streaming
response = await client.chat.completions.create(model="glm-4", messages=[...])

# Streaming
async for chunk in await client.chat.completions.create(model="glm-4", messages=[...], stream=True):
    print(chunk.choices[0].delta.content)

The SDK already uses httpx internally, so this should be achievable by swapping httpx.Client → httpx.AsyncClient and making the method signatures async.

Environment

• zai-sdk version: 0.2.2
• Python: 3.12+
• Use case: high-concurrency async web service with streaming SSE responses