# Actual Computer

> Actual turns hardware you already own into a private AI inference
> cluster. Install one binary, authorize the device against an Actual
> account, and you get OpenAI-compatible and Anthropic-compatible APIs —
> served locally from the machine itself, or remotely through Actual's
> end-to-end Private Relay at api.actual.inc.

This document is for AI agents. If a human gave you "actual.inc" and asked
you to set up local inference, follow the runbook below in order.
Registration is not optional: an install only works once a human has
authorized the device against an Actual account.

## Step 1 — Install

macOS / Linux:

    curl -fsSL "https://actual.inc/install" | bash

Windows:

    powershell -c "irm https://dl.actual.inc/install.ps1 | iex"

Then start it:

    actual

## Step 2 — Register and authorize the device (REQUIRED)

On first run, `actual` starts a device-authorization flow and prints a
verification URL of the form:

    https://actual.inc/device?code=XXXX-XXXX

You (the agent) MUST relay this exact URL and code to your human and wait.
They open it in a browser; if they do not have an Actual account yet they
will be sent to https://actual.inc/register to create one (email + password), after
which the device is authorized automatically. The code expires after 5
minutes — if it expires, run `actual` again for a fresh one. Never invent
an email address or attempt to complete registration yourself.

There is no anonymous mode. No authorized device, no working instance.

## Step 3 — Run inference

Local (same machine, no per-request auth on loopback):

    Base URL: http://127.0.0.1:8080
    POST /v1/chat/completions   (OpenAI Chat Completions-compatible)
    POST /v1/responses          (OpenAI Responses-compatible)
    POST /v1/messages           (Anthropic Messages-compatible)

The server returns 503 until a model is loaded; load one first (Step 4).

Remote, via Actual's Private Relay (requires an API key the human creates
at https://actual.inc/user/keys):

    Base URL: https://api.actual.inc
    Authorization: Bearer sk_...
    Optional: X-Cluster-ID header to pin a specific cluster.

The relay is end-to-end: Actual cannot read prompts or responses.

## Step 4 — Models

    actual models search <query>     # find a model
    actual models download <model>   # fetch it
    actual models list               # see what is installed

Public registry (human-browsable): https://models.actual.inc
Machine-readable index: https://models.actual.inc/index.json

## Links

- Register: https://actual.inc/register
- Model registry: https://models.actual.inc
- Machine-readable model index: https://models.actual.inc/index.json
- Full agent guide: https://actual.inc/llms-full.txt
- Docs (require a signed-in account): https://actual.inc/user/docs/getting-started