A personal assistant desktop app that runs on your own computer. No AI key, no account, no cloud.
- Daily briefing: see today's reminders, tasks and weather when you first open the window.
- Weather: ask about any day in the next week, here or anywhere else.
- Reminders: get a Windows notification at the time you ask for.
- Tasks and notes: add them by typing, and search your notes later.
- Network (optional): check your connection, see what's on your network, and get told about outages and new devices.
- Local AI (optional): a model on your own computer handles what the commands don't understand.
- Talking (optional): have a spoken conversation, entirely on your own computer.
- Camera (optional): hold something up to your webcam and ask about it, entirely on your own computer.
OSSI only understands what it's been taught, but it copes with typos, polite wording and follow-ups
like finish it.
Your data stays in a local file, ~/.ossi/ossi.db, outside this repo. OSSI only uses the internet
for weather, testing your connection, downloading the optional add-ons, and checking for updates
when you ask. Your tasks,
notes and reminders are never sent.
You need Windows and Python 3.11 or newer. There's nothing else to install, unless you want the optional Local AI, Talking or Camera.
Download the latest release, unzip it,
and double-click OSSI.pyw to open the app. The first time, OSSI offers to add itself to your Desktop
with its own icon.
If you delete the shortcut, or OSSI moves, you can add it again in Settings. A shortcut made by hand shows Python's icon instead.
Type in the box at the bottom of the window and press Enter. Type help to see every command.
| Type | What happens |
|---|---|
briefing |
Shows your daily briefing |
set my location to 02108 |
Sets your location, by ZIP code or city |
weather or will it rain tomorrow |
Shows the weather for your location |
what's the weather in Paris |
Shows the weather somewhere else |
remind me tomorrow at 3pm to call the bank |
Sets a reminder |
reminders |
Lists upcoming reminders |
cancel reminder 2 |
Cancels a reminder |
is my internet working |
Checks your connection |
what's on my network |
Lists the devices on your network |
add task call the bank |
Adds a task |
tasks |
Lists open tasks |
done 1 or finish buy milk |
Completes a task, by number or name |
undo |
Reopens the task you completed most recently |
note the wifi password is on the router |
Saves a note |
find note wifi |
Searches your notes |
exit |
Quits OSSI completely |
You can also click the circle next to a task to complete it, and right-click a message to copy it.
Click Settings at the bottom of the sidebar to turn on the AI or the camera, set your location, switch between °F and °C, turn off the daily briefing, change the accent color, start OSSI when you log in, add it to your Desktop, report a bug, or sponsor OSSI.
Turn on Watch my network in Settings and OSSI checks your connection every minute while it's running. It tells you about outages, slowdowns and new devices. It can only watch, not change your router's settings, but Open Google Home in the sidebar takes you to where you can. Phones can use changing addresses, so one phone may show up as new more than once.
Reminders only fire while OSSI is running, so closing the window hides it instead of quitting. To
stop it completely, click Quit OSSI at the bottom of the sidebar or type exit. To
keep them working after a restart, turn on Start when I log in in Settings. Anything that came
due while OSSI wasn't running appears as a missed reminder the next time it starts.
In Settings, click Check for updates, then Update now. OSSI restarts on the new version and your tasks, notes and settings stay safe. Anything typed in the box and not yet sent is lost.
OSSI only checks when you ask. If you got OSSI with git, use git pull instead and open it again.
OSSI works fully without AI. To have it handle messages like
I need to call the bank sometime tomorrow, run a model on your own computer with
Ollama. Installing Ollama doesn't download a model, so that takes a separate
command:
-
Download Ollama from ollama.com and install it.
-
Open Command Prompt: open the Start menu, type
cmdand press Enter. -
Paste this command into Command Prompt and press Enter. It downloads the model, which is about 2 GB, so it can take a while.
ollama pull llama3.2If it says
ollamaisn't recognized, close Command Prompt, open a new one and try again. -
When it says
success, go to OSSI, click Settings and turn on Use a local AI model.
While the AI is on, OSSI starts Ollama for you if it isn't already running.
Good to know:
- Privacy: your messages stay on your computer. The model sees your message, the last few messages, the time, and your open tasks and upcoming reminders. It doesn't see your saved notes or location, unless they came up in those recent messages.
- Control: OSSI asks before running anything the model suggests, and the model can't change your settings.
- Speed: replies appear as the model writes them, and the first words can take a while without a
dedicated graphics card. Say
never mindto stop one. - Accuracy: models can get things wrong, so check what it suggests.
Talk to OSSI and it talks back, for as long as you like.
- Set up the Local AI first.
- Click Talk next to the message box. The first time, OSSI offers to install its voice add-on, about 600 MB. After that it works offline.
- Just start talking. A glow shows what OSSI is doing and reacts to both your voices. Say
goodbyeor click End to finish.
Good to know:
- Use a headset. You can talk over OSSI to interrupt it, but on speakers the microphone can hear OSSI's own voice and cut it off.
- Privacy: the microphone only stays on during a conversation, and you can click Mute at any time. What you say is never saved as audio or sent anywhere, though the words appear in the chat.
- Speed: expect a pause of a few seconds before each reply.
- Actions: looking things up happens right away. For anything that changes something, like adding a task, OSSI asks first and you answer out loud.
- Memory: OSSI forgets the conversation when it closes.
- Language: English only for now.
To hear and choose a voice, or to remove the add-on, go to Settings.
Show something to your webcam and ask about it, like what am I holding, read this label or
what color is this.
- Plug in a webcam and install Ollama.
- In Settings, turn on Let OSSI use my camera. The first time, OSSI downloads a vision model and a small add-on, about 3.5 GB. After that it works offline.
- Click Camera next to the message box for a live view, and line up your shot.
- Type a question and press Enter, or just press Enter for a description. OSSI shows the picture it took, then answers. Click Close when you're done.
You can also ask without opening the live view, and OSSI takes one picture. The live view is mirrored to make lining things up easier, but the pictures aren't, so text in them reads normally.
By voice: during a conversation, say open the camera, then ask about what
you're showing it, and say close the camera when you're done. With the live view closed, you can
still say what am I holding.
Good to know:
- Privacy: the camera is only on while the live view is open, or for a moment when you ask. It closes itself after 2 minutes without a question, and whenever you open Settings or close the window. Pictures aren't saved or sent anywhere. Each one stays in the chat, in memory, until OSSI closes.
- Control: the AI model can't turn on the camera. Out loud, OSSI asks first and you answer yes or no, unless the live view is already open.
- What goes to the camera: with the live view open, anything you type that isn't a command is a
question about the picture. Out loud, only what sounds like it's about the picture is, like
what brand is it, and the rest stays ordinary conversation. - Speed: without a dedicated graphics card, expect up to a minute for each answer. It needs about
6 GB of free memory. Say
never mindto stop one. - Accuracy: models can get things wrong, so check anything important, like a medicine label.
To turn it off or remove the add-on, go to Settings.
See docs/development.md for terminal commands, the project layout, adding a skill and running the tests.
See LICENSE.


