Intelligence, Engineered.

Your own AI stack. Your hardware, your model, your memory.

Prometheus is the agent daemon that runs on hardware you own. OAra Beacon is the cockpit that sits on it — on your desk, and in your pocket.

First time? Get started →

Nothing moves unless something is true.

Honest by construction

“Connected” is a claim. Beacon only makes it when it can prove it.

The link proves itself every twenty seconds. Activity moves only when a real daemon frame lands. A 401 is reported as a 401, and a failed send is never swallowed.

Read the trust story →
Beacon's Mission home: the Armilla telemetry sphere on the left, and on the right the Mission Control panels for agent state, background jobs, models, telemetry and activity.
Mission home — the Armilla beside Mission Control.
Beacon iOS on its Status screen: the same Armilla, a green ONLINE chip, the daemon idle.
Beacon iOS · private beta

Security

What’s enforced, and where it stops

Each line is a claim the Prometheus page makes, with its limits left in, and links to the detail.

A gate that says no

Tool calls pass a security gate with four trust levels: blocked, approve, auto and autonomous. Your deny lists sit over a floor no mode waives: rm -r at a protected root is refused in every mode.

→

A kernel floor on Linux

Bash’s writes stop at the workspace, held by the kernel in a bubblewrap mount namespace and decided on the path the kernel resolves, not on the command string. By default the floor is attempted and says so on every call where it’s missing. Set it to required and bash refuses to run rather than fall back to an unconfined shell. There is no macOS equivalent.

→

Nothing phones home

No telemetry upload, no analytics, no update check. What does leave is listed one line per component: prompts to the model you configure, web search and fetch (on by default), and the chat apps you connect.

→

Tokens redacted

Known token shapes become <redacted> before conversation history, memory, telemetry or training data is written.

→

Coding runs, clean environment

A coding run’s subprocess starts with an empty environment plus at most six variables (PATH, HOME, LANG, LC_ALL, TZ, TMPDIR), so your provider keys never reach it. That covers the environment, not the key file on disk.

→

Messages carry their source

Every message is tagged with where it came from, and a background task’s result comes back fenced as data, not instructions. Web pages, file contents and other tool output aren’t fenced.

→

Pairing and tokens

A new Beacon joins when you approve it, after checking that both screens show the same code, and gets its own token. It works over localhost or your tailnet. The control plane is token-authenticated, and a new install listens on localhost only; a config with no bind set listens on every interface.

→
The full detail, on the Prometheus page →

The shape of it

How it fits together

How Prometheus, Beacon and the rest fit togetherYou talk to Prometheus through Beacon on your desktop, Beacon on iOS (private beta), or a chat app: Telegram, Slack or Discord. Prometheus, at the center, runs on hardware you own, with its tools, its memory and its permission gate. Underneath it are the model it thinks with, local (llama.cpp or Ollama) or in the cloud, plus MCP servers and web tools. On one side, the pillars Instinct, Cognition and Blackboard are being designed; on the other, the extension OAra Voice is coming soon. Solid lines ship today; dashed ones don't yet. ships today designing or coming soon You Beacon desktop · shipping iOS · private beta Chat apps Telegram · Slack · Discord shipping Where you talk to it: on your desk, on your phone, or in a chat app you already use. Prometheus shipping · open source tools memory permission gate runs on your hardware Your agent, on a computer you own. Its memory stays there, and it works within the permissions you set. Pillars Instinct Cognition Blackboard designing Extensions OAra Voice two-way voice coming soon Being designed. Add-ons for Prometheus. Models local: llama.cpp, Ollama or in the cloud MCP servers tools you add Web tools search · fetch The model it thinks with, on your machine or in the cloud, and the tools it can use. How Prometheus, Beacon and the rest fit togetherYou talk to Prometheus through Beacon on your desktop, Beacon on iOS (private beta), or a chat app: Telegram, Slack or Discord. Prometheus, at the center, runs on hardware you own, with its tools, its memory and its permission gate. Underneath it are the model it thinks with, local (llama.cpp or Ollama) or in the cloud, plus MCP servers and web tools. On one side, the pillars Instinct, Cognition and Blackboard are being designed; on the other, the extension OAra Voice is coming soon. Solid lines ship today; dashed ones don't yet. ships today not yet You Beacon desktop · shipping iOS · private beta Chat apps Telegram · Slack · Discord shipping Where you talk to it: on your desk, on your phone, or in a chat app you already use. Prometheus shipping · open source tools memory permission gate runs on your hardware Your agent, on a computer you own. Its memory stays there, and it works within the permissions you set. Pillars Instinct Cognition Blackboard designing Extensions OAra Voice two-way voice coming soon Being designed. Add-ons. Models local: llama.cpp, Ollama, or the cloud MCP servers tools you add Web tools search · fetch The model it thinks with, on your machine or in the cloud, and the tools it can use.

Your hardware

Where it runs

Beacon is the app you talk to. Prometheus is the assistant itself. They can share one computer or live apart.

Simplest start

One Mac

Prometheus and Beacon on the same Mac with Apple silicon. Beacon sets Prometheus up for you.

Home network or tailnet

A computer you own

Prometheus on an always-on Mac or Linux machine at home, with Beacon on your laptop, connected over your home network or your tailnet.

A new install listens only on its own machine, so you set web.bind to reach it from your laptop: how ↗

Private beta

Your phone

Beacon for iOS connects to that same Prometheus, through a desktop Beacon already paired with it. It’s in private beta, not on the App Store yet.

Models are a separate choice: a cloud provider, or local models on your own GPU machine.

The stack

Today, and next

CoreShipping · private beta
Live · open source

Prometheus

The agent daemon. Runs on hardware you own.

Works with Telegram · Slack · Discord · MCP servers

Live · open source

Beacon

The desktop cockpit. Sits on Prometheus.

Private beta

Beacon iOS

The cockpit in your pocket. Same daemon, same control plane.

Designing

OAra Instinct

Fast decisions. Which model, which tool, when to stop.

Designing

OAra Cognition

Planning, and checking the work against the goal before delivering it.

Designing

Blackboard

Coordination across several machines.

Coming soon

OAra Voice

Two-way voice conversation with your agent, coming soon. (The full install can already speak replies, in Telegram and from the command line.)

Get Beacon

Free · desktop app open source (MIT). Pairs with your daemon.

Desktopavailable

macOS

Download↓

Apple silicon. Signed and notarized by Apple.

Desktopavailable

Linux

Download↓

x86_64 AppImage or .deb. Tested on Ubuntu 24.04 · unsigned.

Mobilebeta

iOS

Join the beta→

Private beta. Same daemon, same control plane.

Daemonavailable

Prometheus

Install→

From PyPI, Homebrew or Git, then run the setup wizard.

New to Beacon? Get started, screen by screen →