Production AI, on our cloud or yours

Build AI. Run AI. Scale AI.

Build with one OpenAI-compatible API for text, vision, audio and video models from many providers. Run each request on the model that fits it: a lean model for a quick prompt, a frontier model for a hard one, another provider when one slows down. Scale on GPUs and dedicated endpoints, and pay as you go at live prices.

Start on OneInfer Cloud today, or have our engineers run the same stack in your own cloud or data center with OneInfer Private Cloud, released in stages. Want a finished product? Try OneInfer Voice, Studio and Aura.

Start building Explore Private Cloud

OneInfer Private Cloud: Our engineers · Your cloud or data center
OneInfer Cloud: Model APIs · Routing · Kernel Forge · GPUs
Products: OneInfer Voice · Studio · Aura

OneInfer Private Cloud. Run AI inside your own cloud or data center, built and run with our engineers. The platform is released in stages.

Explore OneInfer Private Cloud

Run AI inside your own cloud or data center

OneInfer Private Cloud brings the same stack into your cloud account or data center: on models you control, at a cost you can plan and charge back.

Our engineering team is available today. OneInfer Private Cloud is in development and released in stages.

Explore OneInfer Private Cloud

Build on OneInfer Cloud today

Our managed AI cloud: one API for text, vision, audio and video models, with routing across providers and kernels tuned to your workload.

Routing that reads every request
Short prompts take lean, low-latency models; hard questions go to frontier models; traffic moves when a provider stumbles.
Kernel Forge
Agents trace your hottest path, generate Triton and CUDA candidates, and keep the fastest.
The SDK you already use
Point the OpenAI SDK at OneInfer with your API key, then call any model in the catalog.

Explore OneInfer Cloud

Call any model with the OpenAI SDKTypeScript

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://api.oneinfer.ai/v1',
  apiKey: process.env.ONEINFER_API_KEY,
});

const response = await client.chat.completions.create({
  model: 'openai/gpt-4.1',
  messages: [{ role: 'user', content: 'Write a blog post' }],
});

How a request runs on OneInfer

You write one call. OneInfer checks it, routes it, runs it on tuned kernels and GPUs sized to the load, and streams the answer back.

Send one request

Point the OpenAI SDK you already use at OneInfer and call any model in the catalog by its id. Ask for a stream and the answer arrives token by token.

TypeScriptOpenAI SDK

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://api.oneinfer.ai/v1',
  apiKey: process.env.ONEINFER_API_KEY,
});

const stream = await client.chat.completions.create({
  model: 'openai/gpt-4.1',
  messages: [{ role: 'user', content: 'Explain DNS briefly' }],
  stream: true,
});import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://api.oneinfer.ai/v1',
  apiKey: process.env.ONEINFER_API_KEY,
});

const stream = await client.chat.completions.create({
  model: 'openai/gpt-4.1',
  messages: [{ role: 'user', content: 'Explain DNS briefly' }],
  stream: true,
});

Response, illustrationDone

DNS turns names like oneinfer.ai into the IP addresses computers use to reach each other.

OneInfer checks and routes it
Guardrails check every request first. Routing then picks the model and provider that fit: lean, low-latency models for short prompts, frontier models for hard questions, and another provider when one stumbles.
Illustration: a request passes the guardrails gate inside OneInfer, then routing sends it down one of three paths. A short prompt goes to a lean model, a hard question goes to a frontier model, and when a provider is slow the request goes to another provider.
One request, checked and routedIllustration
- Short promptA lean, low-latency model
- Hard questionA frontier model
- Provider slowAnother provider takes over
Kernels tuned to your GPU
Kernel Forge traces the hottest path, generates Triton and CUDA kernel candidates, benchmarks them and keeps the fastest correct ones, on GPUs sized to the load.
Illustration: five kernel candidates race. The fastest one fails its correctness check and drops out; the fastest correct candidate is kept and slots into the GPU. Kernel Forge is in preview in the Console as SoL Optimizer.
Benchmarking kernel candidatesIllustration
- Triton ACorrect
- CUDA BFastest correct, kept
- Triton CFails check, dropped
- CUDA DCorrect
- Triton ECorrect
Bar length is relative run time: shorter is faster.
Your GPURunning the kept kernel
CUDA B
Kernel Forge, in preview in the Console as SoL Optimizer.
The answer comes back with one trace
The response streams back to your code, with one trace for the whole request, from the guardrails check to the last token.
Illustration: a trace fills in span by span: guardrails, routing, the model, kernels on the GPU, and the streamed response, which arrives last.
One trace for the whole requestIllustration
- Guardrails
- Routing
- Model
- Kernels on GPU
- Response
- Request inLast token out
Response streamed back to your code
DNS turns names like oneinfer.ai into the IP addresses computers use to reach each other.

TypeScriptOpenAI SDK

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://api.oneinfer.ai/v1',
  apiKey: process.env.ONEINFER_API_KEY,
});

const stream = await client.chat.completions.create({
  model: 'openai/gpt-4.1',
  messages: [{ role: 'user', content: 'Explain DNS briefly' }],
  stream: true,
});import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://api.oneinfer.ai/v1',
  apiKey: process.env.ONEINFER_API_KEY,
});

const stream = await client.chat.completions.create({
  model: 'openai/gpt-4.1',
  messages: [{ role: 'user', content: 'Explain DNS briefly' }],
  stream: true,
});

Response, illustrationDone

DNS turns names like oneinfer.ai into the IP addresses computers use to reach each other.

Put our AI products to work

Ready-made apps from the OneInfer team, each on its own site.

OneInfer Voice
voice.oneinfer.ai
Voice agents that pick up on the first ring.
Answers, understands and resolves calls for enterprises at call-center scale, with natural interruptions and replies that start before the sentence is finished.
Try OneInfer Voice
OneInfer Studio
studio.oneinfer.ai
A generative video, audio and cinema suite.
Create video and audio with generative models, in one studio.
Open Studio
Aura
aura.oneinfer.ai
An AI that answers your website's visitors.
It answers from your own pages, shows you who's on the site now, and hands the chat to your team when a person is needed. By chat or by voice, on your OneInfer credits.
Explore Aura