Can Axis run with a model hosted on our own network?

Yes. Any OpenAI-compatible endpoint works, including vLLM, Ollama and LM Studio on your own hardware.

What keeps working while the model is offline?

The dashboard, the REST API and the MCP interface all keep working. The assistant resumes when the model returns.

How are updates applied?

A new release boots, applies its migrations and then serves. Your data stays in your database volume throughout.

Can we use one model for everyone?

One endpoint serves the whole deployment, so every seat uses the same model.

Can people connect their own AI tools?

Yes. Each person creates a key in Settings, and an MCP client reaches Axis at your API address followed by /mcp. Administrators see and revoke every key in Users and Roles, and one setting on the API switches keys off for the whole deployment.

A Fish Audio key and agent, set as two settings on the API. Talk appears in the assistant once both are in place.

Deployment

Your server. Your model.

Axis runs as three containers on infrastructure you choose, and the assistant connects to the model you configure, including one you host yourself.

Arrange a walkthrough

On your server

Three containers, one deployable.

A PostgreSQL database for the ledger, the API, and the web dashboard with its documentation. The schema sets itself up on first boot, and every boot applies pending migrations before it serves.

Database: PostgreSQL
API: the action layer, the assistant and document reading
Web: the dashboard and the documentation

Your model

Any OpenAI-compatible endpoint.

Point Axis at a model with three settings: an address, a key and a model name. Self-hosted endpoints such as vLLM, Ollama and LM Studio work, as does a gateway in front of several models. Document reading uses the same connection.

What to look for in a model:

Reliable tool calling
A context window of about 32k or more
Close instruction-following

LLM_BASE_URL=https://your-model.internal/v1
LLM_API_KEY=your-key
LLM_MODEL=your-model-name

Changing model means changing three settings and restarting. The action catalogue stays the same.

Your data

Your data stays where you put it.

The ledger lives in your database. The assistant sends the results of the actions it runs to the model you configure, so a model on your own network keeps that data on your network too. Two channels reach further: Talk, through Fish Audio, and any AI tool a person connects with a key.

Ledger: your database
Assistant context: your model
Talk: Fish Audio
A connected AI tool: its own provider

What you need

A server, a database and a model.

A Linux server running Docker, or a host for a direct install with systemd and nginx
PostgreSQL for the ledger
A model endpoint hosted or your own
A domain for the dashboard and the API

How it starts

From first call to first approval.

01
Walkthrough.

We show Axis with your kind of documents and agree the seats and modules.
02
Install.

Axis goes onto your server, connected to your database and your model.
03
Set up.

Your company, chart of accounts, roles and opening balances.
04
First approvals.

Your team reads their first documents and approves their first postings.

Questions

Yes. Any OpenAI-compatible endpoint works, including vLLM, Ollama and LM Studio on your own hardware.

Have a different question? Write to Saad.Khalid@autosphere.ai