Argus
Argus desktop assistant

AI that understands.
And takes action.

Argus is your personal AI assistant for desktop.Talk by voice or text, research, work with files and turn source material into editable presentations.

Release statusGitHub repository
Clone the source code
$git clone https://github.com/reenji/Argus.git
Versionv0.1.0LicenseMITv0.1.0 released • installers available
Stylized Argus eyeA geometric amber eye with a softly pulsing core.
ARGUS / v0.1.0
Gemini Live
Personal desktop AI assistant
Capabilities

Capabilities documented in Argus.

The repository includes a desktop app, optional local tools and a web service with a Next.js client. Integrations depend on credentials, permissions, dependencies and installed applications.

01

Voice and text with Gemini Live

Talk by voice or send text instructions in the desktop app.

Gemini Live requires a Google API key and internet; voice also requires a microphone.
02

Web search and Deep Research

Search the web or run investigations that gather and compare evidence from multiple sources.

Reports cite sources and are written to disk only when you ask to save them.
03

Files and coding

Ask Argus to summarize PDFs, extract text from images, analyze CSV files and help with code.

These examples are in the usage guide; other actions may require optional dependencies.
04

Screen, camera and computer control

Analyze the screen or camera, browse the web, open applications and perform local actions.

Availability depends on the operating system, permissions, dependencies and installed applications.
05

PowerPoint presentations

Create, edit, redesign and extend editable widescreen .pptx presentations.

Sources include PDF, Word, PowerPoint, Excel, CSV, JSON, Markdown, text, images, audio, video and URLs. PDF export requires Microsoft PowerPoint.
06

Reminders, email and messages

Set reminders, read and prepare emails, and work with messages in supported applications.

Gmail uses OAuth; prepared emails are sent only after explicit approval. Messaging depends on available apps and permissions.
Quickstart

Install from source.

The desktop app runs on Windows, macOS and Linux. Docker Compose starts the FastAPI service and Next.js client locally; this profile does not open the desktop UI.

Linux

Desktop app installed from source; launcher: scripts/start_argus.sh.

Install commandsBash
$ git clone https://github.com/reenji/Argus.git Argus cd Argus python scripts/setup_argus.py
RequirementsPython 3.11+, internet and a Gemini key with Live API access; microphone for voice.
Other optionsStart with scripts/start_argus.sh.
01
Clone and set up the environmentRun python scripts/setup_argus.py. It creates .venv, installs dependencies and copies .env.example to .env if needed. Git and Python 3.11+ are required.
02
Set your Gemini keySet GEMINI_API_KEY in .env. Gemini Live requires a Google key with Live API access.
03
Start ArgusUse your operating system launcher in scripts/, or activate .venv and run argus. A microphone is needed for voice conversations.
Sizing

Executed benchmark results.

Automated run on 2026-10-20 in GitHub Actions: SQLite API, web via Docker Compose with PostgreSQL/Redis, and idle desktop. Metrics depend on the runner and are not a production guarantee.

DESKTOP / MEASUREMENT

Desktop app — idle

GitHub Actions Linux x86_64 runner, 60 s, 60 samples: mean CPU 15.851% of one logical core.

Process RSS: 136.73 MiB mean and 136.914 MiB maximum. Typical and intense scenarios were not run because they require real interaction.

Dedicated GPU: not listed as a requirement. Generative inference uses the Gemini API.

The repository publishes no storage minimum; the desktop benchmark did not measure disk usage.

LOCAL API / TEMPORARY SQLITE

Local HTTP API — SQLite

1, 5, 10, 20 and 40 virtual sessions; all levels completed with 0% errors. p95: 2.437 ms, 5.656 ms, 6.951 ms, 24.361 ms and 45.545 ms, respectively.

TEMPLATES / FLY.IO + RENDER

Docker Compose — PostgreSQL + Redis

1, 5, 10, 20 and 40 virtual sessions; all levels completed with 0% errors. p95: 4.873 ms, 9.513 ms, 21.320 ms, 76.874 ms and 146.740 ms, respectively.

Run on a GitHub Actions x86_64 runner with 4 logical CPUs and 15.61 GiB RAM. Results vary with hardware, environment and workload.

Technical reports and artifacts from the 2026-10-20 run:

GitHub Actions run ↗API / SQLite report ↗Docker Compose report ↗Desktop report ↗
Releases

Portable executable or source code.

Version 0.1.0 is published on GitHub. The release includes installers for Windows x64, macOS arm64 and Linux x86_64, plus SHA256SUMS.txt.

GITHUB / RELEASESMIT

Download Argus for your system.

Release v0.1.0 includes installers for Windows x64, macOS arm64 and Linux x86_64. Check SHA256SUMS.txt before installing.

How to start after downloading

Windows x64: open the downloaded .exe file.

macOS arm64: open the .dmg file and drag Argus to Applications. The first launch may require approval in Security settings.

Linux x86_64: install the .deb package with your distribution's package manager. System dependencies may be required.

Follow the GitHub release
Release statusVersion v0.1.0 was published on 2026-10-07 with installers and checksums for all three supported systems.The release includes SHA256SUMS.txt with hashes for the distributed files.
Architecture

Desktop app, web client and Gemini Live.

01InputVoice, text, files, screen and camera
02PyQt6 desktop or Next.js web clientDesktop app and web client
03ArgusPersonal AI assistant
04ToolsLocal tools and optional integrations
05ModelGemini Live through the Google API
INPUTTOOLSRESPONSE
Example workflow

A day of tasks, from the first prompt to review.

An illustrative flow using files, research, presentations and email. Messages and times are examples; emails are sent only after approval.

  1. YOUSummarize these documents and highlight what changed.
  2. ARGUSSummary ready, with references to the source files.
  3. YOUCompare these options and research what is missing.
  4. ARGUSI compared the sources; citations are ready for review.
  5. YOUTurn the report into an editable presentation.
  6. ARGUSPresentation draft ready for review.
  7. YOUPrepare an email with the summary.
  8. ARGUSDraft ready. I will send it only after your approval.
Security

Where the app runs affects where data is stored.

◌

Data and credentials

On desktop, memory and settings are stored in the local Argus data directory. The Gemini key is remembered only if you choose that option; it uses the system keychain when available. The web service stores data in PostgreSQL and encrypts Gemini keys.

◎

External services and telemetry

The audited code includes no analytics, product telemetry or crash-report uploads. Voice, text, files and other information may be sent to Gemini or external integrations when you use those features; see the privacy policy.

!

Permissions and approvals

Screen, file, browser and computer-control tools depend on available permissions, dependencies and applications. Prepared Gmail messages remain visible for review and are sent only after explicit approval.

Questions

Before putting an agent to work.