The official Python client for Akumi, the EU-sovereign,
OpenAI-compatible inference API. Built on httpx, with both a synchronous and an
asynchronous client. One base_url for every model, governed and metered, with
your regulated data kept in the EU.
- Drop-in OpenAI-compatible. Chat completions, embeddings, and models under one key.
- EU-sovereign by default. The egress guard fails closed on non-EU routing.
- Governed, not just hosted. PII firewall, per-request residency, and a metadata-only audit trail on every call.
- Sync and async. Streaming and automatic retries in both clients.
Python 3.9 or newer.
pip install akumiCreate an API key under app.akumi.cloud -> Platform -> API keys:
from akumi import Akumi
client = Akumi.from_api_key("mk_...")
result = client.chat.create(params={
"model": "mistral/mistral-large-latest",
"messages": [
{"role": "user", "content": "Explain EU data residency in one sentence."},
],
})
print(result["choices"][0]["message"]["content"])for chunk in client.chat.create_streamed(params={
"model": "mistral/mistral-large-latest",
"messages": [{"role": "user", "content": "Write a haiku about Frankfurt."}],
}):
print(chunk["choices"][0]["delta"].get("content", ""), end="")import asyncio
from akumi import AsyncAkumi
async def main():
client = AsyncAkumi.from_api_key("mk_...")
result = await client.chat.create(params={
"model": "mistral/mistral-large-latest",
"messages": [{"role": "user", "content": "Hello from the EU."}],
})
print(result["choices"][0]["message"]["content"])
asyncio.run(main())embeddings = client.embeddings.create(params={
"model": "mistral/mistral-embed",
"input": "The quarterly report is ready for review.",
})
vector = embeddings["data"][0]["embedding"]client.models.list()lists the models available to your key.client.memory.forget(...)andclient.memoryThreads.list()manage long-term memory and threads.client.auditLogs.list()reads your metadata-only audit trail.
from_api_key() connects to https://api.akumi.cloud/v1 and retries transient
failures (429, 502, 503, 504). Pass base_url= to target another host.
- Guides: https://akumi.cloud/docs
- API reference: https://akumi.cloud/docs/api-reference
Generated from the Akumi OpenAPI specification, so it tracks the API automatically. Issues: https://github.com/akumi-cloud/python-sdk.
MIT licensed.