Your personal assistant, running on your own computer.
It runs long tasks for you, remembers what matters, notices things before you ask, and talks with you by voice.
Any model, local or cloud. No account, no servers, no setup files.
Get started · Docs · Contribute · Discussions · Roadmap
Note
Sentient v3 is in alpha. It runs from source, its tests run on Windows and Linux on every change, and a Windows installer builds from this repository. Here is exactly what has been tested.
| Everything lives in one folder on your computer. Keys stay in your system keychain. Run it on a free local model and nothing leaves your machine. No account, no telemetry. | Describe a job in plain words. Sentient plans it, asks you once, then runs it now, on a schedule, when an email arrives, or whenever a watched price drops. |
| It remembers what you tell it, builds a picture of your preferences and goals that you can correct, and tidies its memory every night. | Talk to it with "Hey Sentient", message it on Telegram, and pair your phone or smart glasses so it can see, speak and reach you anywhere. |
One-off, recurring, triggered by your apps, or split across parallel helpers. Every run keeps a live log and a
clear report, and failed runs can be retried from where they stopped.
|
"Tell me when this changes" becomes a tiny script that checks on a schedule and only wakes the AI when
something happens.
|
How Sentient understands you, with the reasons behind each thing it noticed. Confirm it, correct it, or tell it
it's wrong.
|
Each night it merges duplicate memories, settles contradictions (an old "lives in Chennai" gives way to
"lives in Bengaluru") and writes a short journal of what changed.
|
Pair a device with a six-digit code. Sentient can show text, speak, check where you are, or take a photo and
tell you what it sees. The protocol is open, with reference glasses firmware.
|
Say "Hey Sentient", talk naturally and interrupt whenever you like. Speech recognition runs on your computer.
|
After real work Sentient writes reusable skills, and proposes fixes when one goes wrong. Nothing changes until
you approve the diff.
|
Pick a model for chat, background work, planning, voice and vision. Mix free local models with your own cloud
keys, and test each one with a click.
|
And more: Gmail, Calendar, Drive, Docs, Sheets, GitHub, Slack, Notion, Trello, Discord, WhatsApp and any MCP server · near-instant updates from Gmail and Calendar, IMAP push and webhooks · proactive suggestions you approve or dismiss · browser control on your own Edge or Chrome · code that runs in a sandbox · helper agents that work beside the chat · messages you can add while it is still answering.
Windows: build the installer with cd desktop && npm run package and run desktop/dist/Sentient-Setup-*.exe.
Prebuilt, signed downloads are coming with the first public release.
From source (Windows, macOS, Linux) with Python 3.12, uv and Node 22:
git clone https://github.com/existence-master/Sentient.git && cd Sentient
uv venv .venv --python 3.12
uv pip install --python .venv/Scripts/python.exe -e ".[voice]" # macOS/Linux: .venv/bin/python
cd desktop && npm install && npm run devThen pick a brain: install Ollama and run ollama pull qwen3:8b for a free, private
model, or paste your own Anthropic, OpenAI or Gemini key. The app walks you through the rest.
Getting started has first things to try and troubleshooting.
flowchart LR
subgraph you["Where you are"]
app["Desktop app"]
phone["Phone"]
glasses["Smart glasses"]
chat["Telegram · Discord"]
end
subgraph engine["Sentient engine, on your computer"]
agent["Agent loop"]
tasks["Tasks and scheduler"]
memory["Memory and user model"]
proactive["Proactivity and skills"]
tools["Tools: apps, browser, code, devices"]
db[("One SQLite file")]
end
local["Local model (Ollama)"]
cloud["Cloud model (your key)"]
apps["Your apps"]
app <--> agent
phone <--> agent
glasses <--> agent
chat <--> agent
agent --- tasks & memory & proactive & tools
tasks & memory & proactive --> db
agent --> local
agent -.-> cloud
tools <--> apps
The desktop app starts a local Python engine. One agent loop powers chat, tasks, helpers, proactivity and voice, and everything is stored in a single SQLite file. Read the architecture, the API contract, and the decision records that explain why it is built this way.
- On a local model, nothing leaves your computer. With a cloud model, only what each request needs goes to the provider you chose.
- Sentient asks before it sends, deletes, buys or runs code. Purchases always ask.
- It never types passwords or card numbers. You sign in to websites yourself.
- Code it writes runs without your keys. Device photos ask first. Only paired chats can message it.
Details in Privacy. Found a vulnerability? Please report it privately: SECURITY.md.
You never need our API keys: Sentient runs on a free local model or your own key, and every test runs offline with a fake model.
- Pick a good first issue or help wanted issue, or fix something that bugs you.
- Fork, branch from
main, and open a small pull request. - Checks run automatically, one maintainer review, and it's in.
Read CONTRIBUTING.md. Using an AI coding agent? Point it at AGENTS.md.
| Getting started | Install, choose a model, first steps, troubleshooting |
| Privacy | What stays on your computer and what leaves it |
| Developing | Run, test, capture screenshots, build installers |
| Architecture · Decisions | How it fits together, and why |
| API contract | Everything the app and the engine exchange |
| Devices protocol | Build your own device or firmware |
| Verification · Roadmap · Changelog | What's tested, what's next, what changed |
Looking for the old cloud version? It lives on the v2 branch.
|
Sarthak (itsskofficial) |
kabeer2004 |
And every contributor. Thank you.
Sentient is free software under the GNU AGPL v3. Contributions are accepted under our Contributor License Agreement.









