Lumo doesn't replace your agent. It keeps thinking, deciding and writing code with its large model, such as Claude Opus. Lumo just takes the reading off its hands. Where every character matters, the agent reads for itself.
Lumo · local AI
Hi, I'm Lumo.
I run on your PC, on the graphics card or in RAM. I don't need a cloud or an account. I take reading work off your AI agent's hands, and I handle small chores on your PC for you.
0conversations with Lumo that leave your PC
−95%tokens on reading tasks I take over (measured)
2.6Mtokens saved, measured across 232 real tasks
15 minwithout work, then I free up the graphics card
The same model, for reading tasks
95% fewer tokens.
Normally, your AI agent reads long logs, large files and test output itself, and every line costs tokens. Lumo reads them locally on your PC and gives back only what's needed, for example the outline, the error lines or the answer to its question.
2.6Mtokens saved, measured across 232 real development tasks
−95%tokens on average for reading tasks Lumo takes over
≈ 11,500tokens saved per task on average
100%the same model, your agent isn't swapped out
Your agent stays
the same powerful model.
Lumo just takes over the reading.
A typical reading task
Average across long test logs, build logs and large source files.
Without Lumo
≈ 12,100 tokens
With Lumo
≈ 600 tokens
Source: Measured with Lumobix across 232 real development tasks with Claude Opus (October 2026): test runs, build logs and large source files. We count what Lumo read instead of the agent, minus what the agent got back. Tokens are estimated at 3.5 characters per token. How much you save depends on your work.
When your agent reads less, its quota with the provider lasts much longer. Its context stays smaller, and even long sessions stay fast and accurate.
Lumobix keeps count on your PC of how often Lumo helped and how many tokens your agent saved as a result. You'll find it in Agent OS under “Helping the agent” and at the bottom left of the chat.
Other ways to save tokens.
There are other ways to make an AI agent more economical. Usually there's a catch: the answers get worse, the tokens are spent elsewhere, or you end up maintaining a handful of plugins. Here's how they compare.
ApproachSaves tokens with the providerSame modelNo extra software
Switch to a smaller, cheaper model✓✗ weaker answers✓
Hand off work to a second cloud agent✗ uses tokens theredepends on the model✗ another account or plugin
Combine several plugins and extensionspartlypartly✗ set up and maintain
Lumo in Lumobix✓ reads locally, no cloud✓ the same model✓ one app, no plugin
Lumobix never secretly switches to a smaller model just to keep going. Your agent works with the model you chose the whole time.
Lumo isn't a second cloud service; it runs on your graphics card. That costs nothing, needs no account, and what Lumo reads stays on your PC. It only passes the parts your agent needs.
Lumo comes with Lumobix. You don't install any plugin or extension; Lumo sets itself up, and your agent can use it right away.
I do the reading, it saves tokens.
Your AI agent gets me as a tool. Instead of reading a huge file or a long log in full, it asks me and gets only the parts it needs. That's faster and saves tokens.
What your agent knows,
as a star map.
Every project is a sun. Every memory is a star, whether it's a decision, a note, an important document or a chat. Lumo sorts them, connects what belongs together and keeps them up to date. That way your agent finds what you've already worked out in seconds and doesn't have to search again.
What your agent remembers, what you note down and which documents matter all end up as stars in the right project.
Lumo sorts new memories, connects related ones and notices what nobody has needed in a long time. All of this happens on your PC.
Instead of searching through old chats and folders, your agent asks the memory. That saves time and even more tokens.
Tell me what to do.
In Agent OS, you write to me like in a chat. I then work directly on your PC, with the tools Lumobix gives me.
What I may do without asking.
You set this under Settings → Lumo. At both levels: I never read keys or passwords, and once you allow something, it applies to the whole conversation.
Standard
- Everyday things without asking: moving files in your user folder, putting them in the Recycle Bin, opening web pages, commands that only display something.
- I ask before touching system folders, AppData or entire standard folders, before starting programs and scripts, and before commands that change something.
Extended Default
- Additionally allowed in your own folders: changing files, starting programs and scripts from there, commands.
- I still ask before system folders (Windows, Program Files and AppData; on the Mac, System and Library), permanent deletion, the registry, installations, services, downloads and ending processes.
I only run commands that reach deep into the system once you confirm them.
If you're gaming, I hold back. After a quarter of an hour without work, I free up the graphics card again.
What you tell me stays on your PC. I don't need an account and send nothing to the internet.
Questions about Lumo.
Do I have to install Lumo myself?
No. Lumo is part of Lumobix. If the AI engine or model is missing, Lumobix sets them up by itself at startup. You can see the progress on the home screen and in Agent OS.
Does Lumo cost anything?
No. Lumo runs on your own hardware, without an account and without a subscription. The models are under their publishers' licenses.
Does Lumo replace my AI agent?
No. Lumo is small and fast and does groundwork for your agent. Planning, programming and changing multiple files remain your agent's job.
Where are the models stored?
In the storage folder you chose during installation. If you later choose a different storage folder, Lumobix sets Lumo up there again.
Lumo is “not available”. What now?
Lumobix tells you the reason: not enough RAM, no 64-bit system or not enough space in the storage folder. Once you've fixed that, click “Check again”. To the guide