Run any model on infrastructure you control.
Xinference is an Australian owned company. We think enterprises should be able to use AI without handing their data, their models, or their bills to somebody else’s black box.
Why we’re here.
Chris Qin started Xinference in 2023 after five years of the same problem: getting code that ran on one machine to run reliably on a thousand. He had built Mars at Alibaba Cloud to do that for pandas and NumPy. When open models arrived, every team that wanted to run one on their own hardware hit the wall he had just spent a company climbing over.
See it in production
What we believe
Your infrastructure, your rules
Customers keep control of their data, their models and their costs. We build for that.
No lock-in
Open models, one OpenAI-compatible API, and models you can take with you. Leaving should always be easy.
Show the numbers
We price the same workload both ways and run the maths with customers before they commit.
An engineer on every account
You talk to the people who build the platform, not a ticket queue.
How we work.
Nothing we ship can cost you speed.
When we published at EMNLP in 2024, the overhead against the raw inference engine measured under 3.64%.
Read the paperWe sit with users before we write the roadmap.
They ask whether a team should keep paying US$15,000 a month for an API or run their own, whether switching from vLLM to SGLang is safe, why a model that runs on a laptop falls over in production.
Talk to an engineerYour data stays where you put it.
Every prompt sent to a closed API leaves your infrastructure.
See dedicated inference- Your models
- Your prompts and data
- Your GPUs, any generation
Sydney and Singapore.
Our engineers and field team are based in Sydney and Singapore, with in-region data residency in both.
See open rolesOpen positions.
Your models and your data should never have to leave your environment for AI to work.
Chris Qin, founder and CEO