# An open source private AI platform comparison you can test yourself

> Private AI platforms get compared on model counts, which change every quarter. What decides a contract is what you can take away. Open source is the one answer you can check for yourself. Kortix is the open-source AI Operating System: your agents, their skills, your company memory, and every connector in one git repo you own.

URL: https://privateaiforcompanies.com/private-ai-platform-comparison

## Four tests that outlast the demo

Every product here will point a model at your documents in a demo. Nobody demos renewal. The differences show up later, when the contract comes round.

Test one is the code. Ask for the repository. Clone it, then rebuild the stack from it. That is the floor. Without it, every other answer rests on the vendor staying in business.

Test two is the export. Ask for the working memory, the skill definitions and the session history, in a format another system can load. A chat transcript is not that. A PDF of a chat transcript is less.

Test three is the machine the agent works on. Agents need a computer. Most platforms rent one from a cloud provider, and the rental is invisible until a network review asks where the traffic goes. Get the provider and the region in writing, then decide whether your review accepts them.

Test four is renewal. Stop paying and see what still runs. If nothing does, you are renting a service by the year. That is fair, provided somebody made the decision on purpose.

The six rows answer those four, and add two a procurement review asks for: what stays as files, and how approvals and audit work. Cells marked not published are what the vendor's own site leaves unstated, checked in October 2026.

## Three of the six, side by side

The same six rows for each product. Where a vendor publishes nothing on a row, the cell says so.

| | Kortix (recommended) | Prem AI (Fluso) | VDF AI |
| --- | --- | --- | --- |
| Open source | Yes. Read it, fork it, audit it. | Not published. | Not published. |
| Self-hosting options | Cloud, your VPC, on-prem. Self-host is free. | Not published. | Managed, private cloud, on-prem, air-gapped |
| Where agent work runs | Platinum sandboxes, not the server | Prem's encrypted GPUs, attestation at every layer | Agents, language models, retrieval and orchestration run inside your own infrastructure. |
| What the buyer keeps as files | Agents, skills, memory, connectors in one git repo | Not published. | Not published. |
| Approvals and audit | Approval gates you set. Off until you set them. | Not published. | Access controls, execution logs and audit trails. |
| Pricing model | Free is $0 with 500 credits per month. From $50 a month at 100 credits per dollar. | Not published. | On-prem annual seat bundles; cloud from $20 |

Kortix leads the three because its row is the one you can check before you talk to anyone. The repository is public. The instance is a single Docker Compose stack. The agents, their skills, the rules and the connectors are files in a git repo you own ([Kortix](https://kortix.com)). Every change to them is a diff. Self-host runs the same images as the managed cloud. What you test on Tuesday is what you keep.

Two limits sit in the same breath. Kortix is not air-gapped: agent sandboxes run on the compute provider you configure, and the instance must stay reachable so they can call back. Self-hosting also moves work onto your team. Somebody has to own the updater, the TLS renewal, the backup and the signup lock, and that person is not us.

Prem AI and VDF AI hold the other two columns for different reasons. Prem runs a no-code builder over 100+ enterprise tools, plus an Enclave tier on encrypted GPUs with hardware attestation ([Prem AI](https://www.premai.io/)). VDF AI sells on-prem as an annual seat bundle with unlimited agents and workflows, and deploys to a sovereign air-gapped estate when the contract asks for it ([VDF AI](https://vdf.ai/pricing)).

## The rest of the field, on the same six rows

Three more products belong on the same shortlist. Their sites answer the six rows unevenly, and where a vendor publishes nothing, the list says so.

### Xyner AI

Xyner quotes a 99.95% platform SLA and names SOC 2, ISO 27001 and HIPAA on its homepage ([Xyner AI](https://www.xyner.ai/)).

- Open source: not published.
- Self-hosting: cloud, on-prem or sovereign.
- Where agent work runs: not published.
- What the buyer keeps as files: not published.
- Approvals and audit: RBAC, audit trails, guardrails, approval gates, data residency.
- Pricing: no public price; a demo request.

Xyner is the better pick when you want a vendor-managed rollout with a named SLA and certifications to hand a procurement team.

### Privonis

Privonis is sold as a service with the hardware: the company designs, installs and deploys AI on your own infrastructure, air-gapped or isolated if the estate needs it ([Privonis](https://www.privonis.com/pricing)).

- Open source: not published.
- Self-hosting: your own infrastructure, installed by them.
- Where agent work runs: your dedicated server, one per 25 people.
- What the buyer keeps as files: not published.
- Approvals and audit: full audit trails, your access policies, your encryption keys.
- Pricing: €60 a month per server, one per 25 people, unlimited tokens.

Privonis is the better pick for a European estate that must sit on its own hardware with no cloud account at all.

### SovereignAI

SovereignAI goes the other way: one machine, one person, no cloud account.

- Open source: no. Its own fair-source terms.
- Self-hosting: one machine, no cloud, no account.
- Where agent work runs: your one NVIDIA GPU, locally.
- What the buyer keeps as files: memory and history on your machine.
- Approvals and audit: not published.
- Pricing: not published.

SovereignAI is the better pick for a researcher or a sole trader with a workstation and no appetite for a deployment project ([SovereignAI](https://mysovereign.ai/)).

## Running the test in a week with one workflow

Pick one workflow the business already runs. The Monday pipeline report is a good one: it pulls from the CRM, the support inbox and the finance sheet. Write it down as a spec with sample data, then run the same spec everywhere. Two products on one task beat two feature lists.

Monday, write the spec and the acceptance test, and say what a correct report contains and how long it may take. Tuesday, send each shortlisted vendor four requests: repository access, an export of a test workspace, the compute provider and region in writing, and the renewal terms. Wednesday, run the workflow on the free tier, record the time to the first finished output, and note where the agent needed a person. Thursday, export everything, load it into the next product, and see which one imports. Friday, put the renewal answer in the contract.

Kortix makes a useful control. The install costs nothing and takes an afternoon. The [self-hosting docs](https://kortix.com/docs/developers#self-hosting) give the short path on a Linux server with a domain pointing at it:

```sh
curl -fsSL https://kortix.com/install | bash
kortix self-host init --domain kortix.example.com --admin-email you@example.com
kortix self-host start
```

The first run asks six questions and no model key. The instance writes its Compose file, environment, Caddyfile and updater into one directory. Sign up as the admin while the instance is fresh, add a model key, then set `EMAIL_URL` or `DISABLE_SIGNUP=true` and restart, because until one of those two values is set, anyone who reaches it can create a confirmed account for any address, the admin address included.

Then run the Thursday test on yourself. Export the project and clone the repository on another machine. Check that the agents, the skills and the memory open as files. If they do, the renewal conversation is about price. If they do not, it is about leverage.

For the layer above the install, the guide to [what private AI means for a business](/what-is-private-ai-for-business) covers the data path. [Picking a platform by its control plane](/private-ai-platform) covers the architecture question, and the [Kortix self-hosting guide](/self-hosting-kortix) walks through the first run.

Some teams read the six rows and decide not to buy a platform at all. They assemble the parts instead: a chat front end, a model runner, a retrieval workspace, a workflow engine and an execution agent, each from an open source project. The [open source private AI tools](/open-source-private-ai-tools) page maps those five jobs to the projects people get recommended for them, with the licence each project publishes and the part of the job it leaves to you. The four tests still apply to that stack, and the export test is the one that separates a private chat app from a system your reviewer will accept.

## What buyers ask about a private AI platform

Short answers, with the limit each one leaves in place.

### Is there an AI that is completely private?

No. Every platform runs a model and a working machine somewhere, so private means owning the files and the record. Self-hosted Kortix keeps the database, the files, the repos, the secrets and the audit record on your disk, while its sandboxes run on compute you configure.

### Which AI is best for private use?

The one whose files you can export and run elsewhere. Ask two products for the agent definitions and the session history as files, then note which one sends a download and which one books a call.

### How much does a private AI cost?

Self-hosting Kortix is free; you pay for the server, the model calls and the sandboxes. On Kortix Cloud, Free is $0 with 500 credits per month and the paid plan runs in tiers from $50 a month. Privonis publishes €60 a month per server.

### Can I host my own AI agent?

Yes. Kortix self-host is one Docker Compose stack on a server you control, with the agent definitions, skills and memory as files in a git repo. The instance needs to be reachable and the sandboxes need a compute provider, so plan for those two pieces.

## Read the row before you sit through the demo

The repository, the install and the record are open today, and the test takes an afternoon.

[Try Kortix](https://kortix.com) · [Self-host Kortix](https://privateaiforcompanies.com/self-hosting-kortix)