Customers
Teams building on Xinference.
From AI-native startups to enterprises with thousands of staff. See what they run and what changed.
AI-native companies
AI-native teams shipping models and agents.
Model API and agents on one stack, from code review to warehouse operations.
Model APITransport and logisticsMoved code review off a closed model. Every push now reviewed in about 10 seconds.
TransVirtual moved its pull request reviews off a closed-model API. Every push into main is now reviewed on DeepSeek V4.1 Flash through Xinference Model API, and a posted review is a required check before merging.
~10 secper pull request reviewclient = OpenAI(
base_url=XINFERENCE_BASE_URL,
api_key="XINFERENCE_API_KEY",
)
# two lines. that's the migration.More AI-native teams
Hosted in AUmodel inference for support agents in the NDIS spaceSupport agents for NDIS providers, on inference hosted in Australia.
AI support agents and an internal assistant, ShiftCare Brain, running on models hosted in Australia.
~18 hrsof stock reconciliation saved a monthEvery cycle count checked by an agent, signed off by a person.
An agent checks every counted line against the warehouse system, with a person approving every change.
Every 5 minnew trials picked up and briefed in SlackEvery new trial gets a sales brief in minutes.
An agent checks HubSpot every five minutes and sends reps talking points and a draft email in Slack.
85%+ lessper task moved to the faster modelSmart models for the hard calls, light models for the rest.
Ads analysis goes to a reasoning model, routine drafts to a faster, cheaper one, behind one API.
< $3per 1,000 chatbot answersA website chatbot that doesn’t need the most expensive model.
Visitors get answers from Cloud Assess’s own content, on a fast, low-cost model.
Enterprise
Enterprises running their AI on one platform.
Dedicated deployments, with the numbers each team reported.
5 days → 2 hrsto put a new model into productionFrom 5 days to under 2 hours to deploy a new model.
6,000 staff use AI for research, document parsing and compliance checks, served by one pooled cluster at 360,000 requests a day.
~30%lower AI costSwapped a self-managed vLLM and Kubernetes stack for one platform.
A regulated AI platform now serves 500K+ requests a day, with new models rolled out in 1 to 2 days.
45 to 65%GPU utilisation, up from under 25%50+ AI use cases across KFC, Pizza Hut and Taco Bell on one shared cluster.
Customer assistants and store analytics run by day, batch work overnight, at 1.52M requests a day.

Xinference gives us one platform to run every model on infrastructure we fully control.

We consolidated model serving onto Xinference and stopped babysitting the stack ourselves.
What customers run
Common workloads on Xinference.
Document processing
Parsing, extraction and compliance checks across large document sets.
Financial services, Yum! BrandsCode review
A review on every push, posted as a required check.
TransVirtualSupport agents
Agents that answer customers and staff from your knowledge base.
ShiftCare, Cloud AssessLead follow-up
A brief on every new lead, in Slack, within minutes.
DetrackOperations agents
Stock monitoring, cycle counts and reordering in the warehouse.
DatapelModel routing
Reasoning models for hard calls, fast models for the rest.
Personify Care, Cloud AssessRun the numbers on your workload.
A cost assessment takes one call. The savings model is yours to keep.