Get Your Hands on
Sovereign
Superintelligence
Own your intelligence.
Actual runs AI on your own computers, at any scale, anywhere.
Our engine runs open-source AI models as fast as each computer allows. Our platform links your computers into one private API, in one data center or across the world.
Runs on everything
Mac, Windows, and Linux computers with NVIDIA, AMD, or Apple chips, all thinking together.
Zero setup
No settings to change. It's tuned for your hardware and stays fast, even with your whole team using it.
Hermetic privacy
Every computation happens on your computer, and our confidential relay can't see what passes through. You get perfect privacy with open internet convenience.
Reach it anywhere
Use your computers from anywhere and invite anyone you choose. The connection stays private.
How it works
Your own AI, as convenient as OpenAI.
Your data stays yours, on computers you own. Our confidential relay keeps it secure all the way there, so Actual can't look into it. Nobody can.
Install
Run one command on each computer. Works on Mac, Windows, and Linux, with NVIDIA, AMD, Apple, and most other chips.
Link
Your computers work together behind one API. Each request goes to a computer that has the model it asks for and room to run it.
Use
Use it for everything: your work, your products, and your research. Agents like Hermes, Codex, and Claude Code plug right in.
For teams
Sovereign AI for your team.
For companies whose data can't leave the building. Actual turns the computers you already own into one private AI service for everyone.
Hardware, chosen with you
We help you pick the right computers for your work and your budget.
- NVIDIA
- AMD
- Apple
- + most other chips
Fully self-hosted, on contract
We set it up where you need it, with your own API endpoint and deeper monitoring of the AI your team hosts.
- On-prem
- Office
- Colocation
- Cloud GPUs
- Shared computers, with a personal key for each person
- One private API for every app your team uses
Actual configures itself for optimal serving on your machines.
On average, users see 2x effective throughput compared to leading inference options.
Questions
- Can Actual see my prompts?
- No, and nobody else can. Your computers do the work, and our confidential relay keeps your traffic secure all the way to them.
- Does it work offline?
- Yes. Every computer running Actual keeps serving AI on localhost, even when it's offline.
- What hardware do I need?
- A Mac, Windows, or Linux computer you already have, from a laptop to a server in a data center. More memory lets you run bigger models.
- Which models can I run?
- Open-source AI models from the Actual catalog. Explore models →
- Is it free?
- Yes! All core functionality is free, including up to 100 million tokens a day through our relay, more than most computers can make. Past that, it's inexpensive. For teams, talk to us.
A computer that can't think isn't an Actual Computer.
A school calculator outguns the machine that flew us to the moon and no one calls it a computer anymore. The bar keeps moving. Actual is how the hardware you already own turns into a thinking computer, with real intelligence, running locally.
