Private by construction
It runs on your hardware and works with no internet connection at all. OfflinAI never receives your prompts, documents or answers.
OfflinAI Server
Your own private AI server.
Run chat, vision, speech, image and embedding models on machines you own, behind the API your software already speaks — with the controls IT needs. Prompts, documents and answers never leave your network.
Or try the Free plan now: run it from Docker Hub.
- OpenAI-compatible API
- Windows and Linux
- Works air-gapped
- Made in Ireland
Your people and apps
Chat app
Office tools
Your software
Agents
OfflinAI Serverthe OpenAI API, at your address
Text and visionon your GPU or CPU
Speech and imagesstarted when needed
Another machinejoined as a remote worker
Models on your own disks
Why OfflinAI
We’d love to use AI, but we can’t put client data into ChatGPT.
Accountants, solicitors, advisers, clinics and public bodies hold other people's information. Most AI services ask them to send it to someone else's servers, often outside the EU.
OfflinAI Server takes that question off the table. The models run on a machine you own — in your office or in your own cloud — and nothing you type, upload or generate leaves it.
Works with what you have
OpenAI's API by default, plus Anthropic's and Ollama's. Tools and programs written for them work once you change one address.
Governed like any other server
API keys, quotas and budgets, single sign-on, a signed audit log and filters that mask personal data — the controls IT and auditors ask for.
One server
Every kind of model, behind one address
Each capability runs in its own supervised worker on your GPUs or CPU. Your people and programs get one address, one set of keys and one usage record.
Chat and agents
Streaming chat with tool calls, JSON output and OpenAI's Responses API.
Vision
Read receipts, invoices, forms, screenshots and photos.
Search your documents
Embeddings and rerankers, the building blocks of private document search.
Speech to text
Transcribe recordings with timestamps, or live audio as it is spoken.
Text to speech
Natural voices, streamed while they speak.
Image generation
Text to image, image to image and inpainting, with optional screening.
Moderation
Guard models check text, and a vision model checks images.
Batches
Large jobs through OpenAI's Batch API, safe across restarts.
How it fits
Up and running in an afternoon
-
Install it on a machine you control
A Windows PC or server, a Linux machine, a container or a Kubernetes cluster. NVIDIA GPUs run through CUDA, other GPUs through Vulkan, or it runs on the CPU alone.
-
Choose your models
Any GGUF model, the models in an existing Ollama store, or downloads from the server's own catalog. Every plan runs any model.
-
Point your tools at it
Chat applications, office tools, your own software and scripts use the API they already speak — each with its own key and its own limits.
Who it's for
Built for people who look after other people's data
Accountancy practices
Use AI on client files without sending them anywhere.
For accountantsLaw firms
Summaries, drafts and search, with privileged documents kept on your own machines.
For law firmsIT providers and MSPs
Offer private AI to your clients, managed like any other server you run.
For IT providersDevelopers
One server, every modality, the OpenAI API — on your own GPUs.
The API
For IT
Everything your security team will ask about
OfflinAI Server is run like any other server: a Windows service, Linux packages or a container, with the controls of an enterprise product from the first release.
Keys, limits and budgets
API keys scoped to endpoints and models, with quotas and budgets per project.
Single sign-on
Microsoft Entra ID, Google, Okta or any OpenID provider, with roles.
Tamper-evident audit log
A signed, hash-chained record, forwarded to your SIEM as it happens.
Content filters
Mask or block emails, phone numbers, card numbers, IBANs and your own patterns.
Encrypted by default
HTTPS with a built-in certificate authority, your own certificates, or ACME.
Observable
Prometheus metrics, OpenTelemetry traces and a live console view.
Air-gapped licensing
Activate online, or carry an offline licence file in and out.
Scale out
Several GPUs in one machine, or other machines joined over mutual TLS.
Pricing
Per server, not per seat
Every plan runs any model you like; plans differ in what the server does around the models. Pay monthly or yearly, or buy a perpetual licence and own it.
Why us
A European company you can hold to its word
- Irish and EU lawOfflinAI Limited is a company registered in Ireland, based in Dublin.
- Your data, never oursYour content stays on your machines. A paid plan only checks its licence — or uses an offline licence file.
- Open standards firstOpenAI's API, OpenID Connect, OpenTelemetry, Prometheus and syslog: nothing ties you to us.
- Plain dealingPublished prices, a readable licence, a security policy and release notes.
Does it need an internet connection?
No. Once your models are downloaded the server runs entirely offline. Licences can be activated online, or offline for air-gapped sites: you carry a small request file to any connected machine and bring the signed licence back.
Does OfflinAI see our data?
No. Prompts, documents and answers stay on your machines. The Free plan sends us nothing. A paid plan checks its licence with our licence service — the licence key, a pseudonymous identifier of the machine, its host name, its operating system and the server's version — or uses an offline licence file instead.
What hardware do I need?
A 64-bit Windows or Linux machine. NVIDIA GPUs run through CUDA and other GPUs through Vulkan; without a GPU the server runs on the CPU. Memory decides which models fit: small models run on a laptop, large ones want a GPU with plenty of memory, or several GPUs.
Will our existing OpenAI code work?
Yes. Set the SDK's base URL to your server and use one of its API keys. The differences from OpenAI's own service are listed in the documentation: for example, speech comes back as WAV, FLAC or PCM rather than MP3, and images as base64 rather than links.
Is there a free version?
Yes. The Free plan runs any model for programs on the same machine, with one engine worker and two API keys. It runs from the container image offlinai/offlinai-server on Docker Hub.
Is macOS supported?
Not yet. OfflinAI Server runs on Windows and Linux, including in containers and on Kubernetes.
Early access
Run AI where your data already lives
OfflinAI Server is in early access. Tell us where you'd like to run it and we'll be in touch with a licence and the packages for your platform.
- A licence and the packages for your platform
- A free 20-minute call to size your hardware
- Founding-customer pricing: 50% off your first year
- A reply from a person, not a sales sequence
Tell us about your setup
Where would you like to run OfflinAI Server, and for what? We will be in touch by email.