Lumen 0.1.6

Tidy your inventory by asking, and undo it if it goes wrong.

Download

Download for Mac	macOS 12 or later, Apple silicon

The Windows installer follows.

Before you install

This software is not provided or supported by Linden Lab, the makers of Second Life. Lumen is
not affiliated with Linden Lab or the Firestorm project. There is no customer support. What it
sends where, and what it does that other viewers do not, is on the
privacy page.

New since 0.1.5

Undo and Redo for your inventory. They are at the top of the inventory window's gear menu,
and each one names what it would undo: "Undo: assistant moved 885 items to Christmas". They
cover the changes made since you logged in, the assistant's and your own moves, renames,
deletes and new folders, up to the last 30. Before anything is put back, the viewer shows you
what will change and asks: Accept or Cancel. You can also just say "undo that".
Anything emptied from the Trash, given away or rezzed cannot come back.

Change many things at once. "Put everything named Christmas into my Christmas folder", "take the
unpack scripts out of my product folders", "delete the folders that are empty now". The
assistant works out a plan first, the viewer asks you before it runs, and the whole change is
one step you can undo. A product in its own folder is moved or deleted whole, never emptied
item by item.

Emptying the Trash. The viewer asks you every time.

Saving the scripts it writes. Only scripts the assistant wrote itself, after the viewer asks
you, and it tells you whether the script compiled.

These last three are off until you switch them on, in Preferences → AI → Permissions
(changing many things and emptying the Trash share one switch). Switching one on shows a short
warning first.

Folders you protect are left alone. Right-click a folder and choose Protect: the assistant
moves, renames and deletes nothing in it. The viewer's own folders (the AO's, the LSL bridge's,
your wearable favourites) are left alone too.

It can run. Ask it to run somewhere instead of walk.

Your unsaved work stays yours. When a window asks "Save changes?", only you can answer it.
And the assistant cannot change the settings that keep it in check: your protected folders, RLV,
or its own permissions.

Before it does anything

The assistant is off until you give it a provider, in Preferences → AI. It can use a
ChatGPT, Claude or Mistral subscription you already pay for, an API key from Anthropic, OpenAI
or Mistral, or a model running on your own computer. Whichever you choose is where your
conversation goes.

Based on

Firestorm 7.2.4.80712.

Lumen 0.1.5

Talk to the assistant instead of typing.

Download

Download for Windows	Windows 10 or 11, 64-bit
Download for Mac	macOS 12 or later, Apple silicon

Before you install

This software is not provided or supported by Linden Lab, the makers of Second Life. Lumen is
not affiliated with Linden Lab or the Firestorm project. There is no customer support. What it
sends where, and what it does that other viewers do not, is on the
privacy page.

New since 0.1.4

A mic button in the Assistant. Click it and talk. What you said is sent as soon as you stop
talking, and no window pops up over the viewer while it listens.

• On a Mac it uses the Mac's own speech recognition. The first time, macOS asks once for
  the microphone and once for speech recognition. English is recognised on your Mac; some
  languages, Danish among them, are sent to Apple to be recognised.

• On Windows it uses Whisper, which runs on your own computer, so what you say never leaves
  it; only the text goes to the assistant, as if you had typed it. The first time you click the
  mic, Lumen asks before it downloads Whisper's speech model (about 190 MB), and you can remove
  it again in Preferences.

  The Windows mic is an early version, and it is not as good as the Mac's yet. It is
  slower: Whisper writes down what you said after you stop talking, so your words appear after
  a short wait rather than as you speak. It can mishear more often, and it understands English
  only for now.

While it listens, your voice chat microphone is switched off, so the people around you do not
hear what you say to the assistant. It comes back on when the mic stops.

In Preferences → AI → Assistant, under Talking instead of typing:

• Language: which language to listen for. It starts out as the computer's own, and lists
  only languages this computer can listen in.
• Send what I say as soon as I stop talking (on). Turn it off and your words wait in the
  box until you press Enter, so you can fix a misheard word first.
• Keep listening after each answer, until I click the mic again (off). One click, and you
  talk back and forth. It stops by itself after two minutes of silence.

Voice chat asks for the microphone when it needs it. On a Mac, Lumen no longer asks for
microphone access the moment it starts. macOS asks the first time you use voice chat or the mic
button.

Clear is a bin icon now, which gives the typing box more room.

Before it does anything

The assistant is off until you give it a provider, in Preferences → AI. It can use a
ChatGPT, Claude or Mistral subscription you already pay for, an API key from Anthropic, OpenAI
or Mistral, or a model running on your own machine. Whichever you choose is where your
conversation goes.

Based on

Firestorm 7.2.4.80712.

Lumen 0.1.4

Mistral joins the providers, your subscriptions work on Windows too, and the assistant can do a good deal more in the world.

Download

Download for Windows	Windows 10 or 11, 64-bit
Download for Mac	macOS 12 or later, Apple silicon

New since 0.1.3

Mistral. Two new choices in Preferences → AI: a Mistral API key, or Mistral Vibe on
the Mistral account you already have, set up for you with Set it up for me… like Codex and
Claude Code. Mistral Medium is the recommended model. Small 4 is faster and cheaper, especially
on a free Mistral plan.

Codex, Claude Code and Mistral Vibe on Windows. Until now Windows could only use an API key
or a local model. Set it up for me… now installs and signs in all three on Windows as well,
with no windows to deal with along the way.

More it can do for you:

• touch an object, and say what clicking it would do instead (sit, buy, pay)
• sit with somebody, on whatever they are sitting on
• offer somebody a teleport, ask them for one, or teleport straight to them
• say where a person is, not only who they are, as far as the viewer can see
• set your group tag, set home, play a gesture, offer friendship
• tidy your inventory: make a folder, move and rename things
• tell you what the parcel's music is, play or stop it, and change it on your own land

Anything that speaks, sounds, or reaches another person asks you first, as before, and
Preferences → AI → Permissions now scrolls, so every one of those questions is there to set.

Fixes from using it. Walking stops when you take over with your own keys and says so;
"take it off" looks at what you are wearing first; a new script appears in one step; and Codex
starts its own background service again after your computer restarts.

Before it does anything

The assistant is off until you give it a provider, in Preferences → AI. It can use a
ChatGPT, Claude or Mistral subscription you already pay for, an API key from Anthropic, OpenAI
or Mistral, or a model running on your own machine. Whichever you choose is where your
conversation goes.

Based on

Firestorm 7.2.4.80712.

Lumen 0.1.3

Lumen and Firestorm now share your AO, and their bridges stop fighting.

Download

Download for Windows	Windows 10 or 11, 64-bit
Download for Mac	macOS 12 or later, Apple silicon

New since 0.1.2

Your #Firestorm folder is left alone. Lumen no longer renames it to #Lumen. Your AO
and the rest stay in #Firestorm and Lumen uses them from there, so switching between Lumen
and Firestorm keeps the same AO in both. Lumen keeps only its own LSL bridge, in a folder
called #Lumen.

If an earlier Lumen renamed your #Firestorm, 0.1.3 gives it its old name back the first
time you log in. If you now have both folders and your AO is in #Lumen, Lumen leaves it
there rather than merge two AOs; to have it in Firestorm too, move the #AO folder from
#Lumen into #Firestorm.

The two bridges no longer fight. After switching viewers, the other viewer's bridge could
stay worn, and each viewer kept rebuilding its own. Lumen now takes Firestorm's bridge off,
and Lumen's bridge takes itself off after about a minute when you log in with Firestorm. Firestorm
may still rebuild its own bridge once, the first time after Lumen.

Before it does anything

The assistant is off until you give it a provider, in Preferences → AI. It can use a
ChatGPT or Claude subscription you already pay for, an API key from Anthropic or OpenAI, or a
model running on your own machine. Whichever you choose is where your conversation goes.

Based on

Firestorm 7.2.4.80712.

Lumen 0.1.2

The assistant now asks before it does anything that matters, can search Second Life
itself, and can run on a model on your own computer.

Download

Download for Windows	Windows 10 or 11, 64-bit
Download for Mac	macOS 12 or later, Apple silicon

New since 0.1.1

It asks first. Before the assistant speaks in chat, sends a message, gives or deletes
something, builds, or teleports you to a place it found, the viewer asks you with a Yes / No
box, and you can tell it to remember your answer. Preferences → AI → Permissions lists
every question with Always allow, Ask and Always refuse. A script asking for permission — to
take your money, say — is never answered by the assistant, only by you.

Also new: places and events from Second Life's own search, and landmarks that know where they
lead ("take me to my hut in Rio Solimoes"). Things around you found by what you call them.
Saving what you are wearing as an outfit. Your notecards read quietly in the background after
you log in, so searching inside them is quick even the first time. A memory for each avatar,
and "remember that…" when you want it to keep something. Settings found by the words printed next to them. Photos framed
properly for small avatars. GPT-6 Sol and Luna through Codex. A local model can have its thinking switched off, and
Preferences → AI → Local says what one needs. The title bar shows the right version, and the check for a
new release works. And 148 defects fixed, found by reading the whole thing through a second
time.

Your #Firestorm folder

Lumen renames your #Firestorm inventory folder to #Lumen; it holds your AO and the LSL
bridge. If you go back to Firestorm, rename it back to #Firestorm first, or Firestorm makes a
new empty one and your AO looks gone.

Before it does anything

The assistant is off until you give it a provider, in Preferences → AI. It can use a
ChatGPT or Claude subscription you already pay for, an API key from Anthropic or OpenAI, or a
model running on your own machine. Whichever you choose is where your conversation goes.

Based on

Firestorm 7.2.4.80712.

Lumen 0.1.1

The first Windows build, with macOS alongside it. Everything else is fixes
and additions since 0.1.0.

Downloads

Lumen-0.1.1-Windows-x64-Setup.exe	Windows 10/11, 64-bit
Lumen-0.1.1-macOS-arm64.dmg	macOS 12 or later, Apple Silicon

New since 0.1.0

A summary of what was waiting for you the first time you open the assistant
after logging in — notices, offers and messages that arrived while you were
away. Answering while away now replies only to the people you named, and tells
them when they arrive rather than only when they write. Whether somebody has a
Primfeed or a Marketplace store. Building in world without the Build window
opening. And twenty-eight defects found by reading the whole thing through.

Before it does anything

The assistant is off until you give it a provider, in Preferences → AI. It
can use a ChatGPT or Claude subscription you already pay for, an API key from
Anthropic or OpenAI, or a model running on your own machine. Whichever you
choose is where your conversation goes.

Based on

Firestorm 7.2.4.80712.

Lumen 0.1.0

The first build of Lumen — a Second Life viewer with an assistant built into
it. You ask for what you want in ordinary words and the viewer does it.

Downloads

Lumen-0.1.0-macOS-arm64.dmg	macOS, Apple silicon

Apple silicon only. There is no Intel build and there will not be one. A
Windows build is being worked on.

The source is here, so anyone who needs a platform that isn't built can make
one.

Before it does anything

The assistant is off until you give it a provider, in Preferences → AI. It
can use a ChatGPT or Claude subscription you already pay for, an API key from
Anthropic or OpenAI, or a model running on your own machine.

Based on

Firestorm 7.2.4.80712.