Argentum

July 26, 2026 · View on GitHub

Argentum is a local-first AI workspace. It runs on your own machine so your data stays with you. You can chat with AI providers you choose, route conversations through Telegram, Discord, or other channels, keep memory across sessions, and use a full desktop app instead of juggling browser tabs.

Website: ag064.eu | Bug reports: GitHub Issues or report@ag064.eu | Privacy and data handling

Install

Quick Start

Download and run (Windows):

Download the Windows setup executable from the latest published release
Run the installer, launch from Start Menu
Follow onboarding, connect a messaging channel, start chatting

Build from source (any platform):

git clone https://github.com/AG064/argentum.git
cd argentum
npm install && npm run build

Windows Setup

For most people on Windows, use the desktop installer:

Download the latest Windows installer

The setup wizard behaves like a normal Windows installer:

shows the license agreement
shows the default install location, usually C:\Program Files\Argentum
lets you change the install folder before installation
adds Argentum to the Windows Start Menu
adds a desktop shortcut
offers to launch the Argentum desktop interface when setup completes

After setup, launch Argentum from the Start Menu. First launch opens onboarding. After that, Argentum opens directly into the desktop app.

v0.0.9 Release

The published v0.0.9 release includes default-deny desktop AI tool/context enforcement, server-side context budgets, Hugging Face GGUF search, bounded local model scan, llama.cpp controls, a safe OpenClaw import/archive flow, help/FAQ feedback, numeric update checks, dependency fixes, startup diagnostics, and release-pipeline corrections. See the v0.0.9 release notes.

The release also contains the optional Argentum llama.cpp local server path. Installers and portable binaries are produced by release automation and attached to the GitHub release.

Provider Status

Release-candidate providers

Provider	Status	Auth modes	Notes
ChatGPT / OpenAI	Candidate	API key, ChatGPT/Codex browser-account authorization	Live chat, model picker, provider test, and usage/error reporting are wired.
MiniMax	Candidate	API key	Live chat and optional Token Plan usage checks are wired; credentials/account combinations still require live validation.
Argentum-managed llama.cpp	Candidate	Local endpoint, optional bundled/in-app binary install	Runs selected third-party GGUF models; there is no custom Argentum model.

Testing Providers

Anthropic Claude, Google Gemini, OpenRouter, NVIDIA, Groq, external Ollama/LM Studio/local endpoints, and Custom endpoints are selectable under Testing. Adapter availability is not a claim that every credential, model, quota, attachment, and tool combination has been live-tested.

Supported OS

Platform	Current support	Assets	Runtime notes
Windows	Windows 10/11 x64	Latest published Windows setup executable and MSI	Uses Microsoft Edge WebView2. Windows 11 includes it; Windows 10 1803+ usually has it, and the installer can install it if missing.
Linux	Source-supported target	Linux desktop packages are planned for release automation	Requires the normal Tauri/WebKitGTK desktop stack. Ubuntu 22.04+ and similar Debian/Fedora/Arch/openSUSE desktops are the intended targets.
macOS	Source-supported target	macOS DMGs are planned for release automation	Uses the system WKWebView. The intended targets are macOS 10.15+ on Intel and Apple Silicon.

Hardware Requirements

Argentum uses the system webview and hosted providers by default, so the desktop app does not need a large local GPU.

Level	Requirement
Minimum	64-bit CPU, 4 GB RAM, 500 MB free disk space, working system webview, internet access for hosted providers.
Recommended	4-core CPU, 8 GB RAM, 1 GB free disk space, stable broadband, and an SSD-backed workspace.
Gateway	Localhost port `3000` available by default, with provider/network access only when you configure and approve it.
Provider usage	ChatGPT/OpenAI usage follows the selected account or API plan. MiniMax Token Plan usage is shown in Diagnostics and reset information is surfaced to the model context.

Portable CLI

Use the CLI when you want the terminal workflow instead of the desktop app. Windows CLI binaries are release assets when built; Linux and macOS CLI binaries can be built from source until CI packaging is enabled.

argentum onboard
argentum doctor

Docker

docker run -it ag064/argentum

From Source

git clone https://github.com/AG064/argentum.git
cd argentum
npm install
npm run onboard
npm run dev

What Argentum Includes

The current project includes:

local configuration and data ownership
73 optional feature source modules at the v0.0.9 review; see the status document before relying on one
Telegram session routing, plus Discord, Slack, WhatsApp, email, SMS, webchat, and mobile channel modules
SQLite, semantic search, knowledge graph, markdown, and hierarchical memory options
encrypted-vault support plus desktop workspace secrets.env, allowlists, policy controls, audit logs, and rate limiting
a desktop shell, a CLI, Docker support, and release packaging scripts

Everyday Commands

argentum onboard                 # Run first-time setup
argentum doctor                  # Check configuration and dependencies
argentum gateway start           # Start the API/web gateway
argentum gateway stop            # Stop the gateway
argentum gateway status          # Show gateway status
argentum tools                   # List available features
argentum feature <name> enable   # Enable a feature
argentum feature <name> disable  # Disable a feature
argentum config                  # Print current config
argentum memory search <query>   # Search memory
argentum help                    # Show all commands

Double-clicking the installed Windows shortcut opens the Argentum desktop interface. The CLI binaries are terminal tools and are intentionally separate from the GUI installer.

Architecture

Argentum
|-- Channels
|   |-- Telegram, Discord, Slack, WhatsApp, email, SMS, webchat, mobile
|-- Agent Runtime
|   |-- LLM providers, model routing, tools, sessions
|-- Features
|   |-- automation, media, integrations, skills, workflows
|-- Memory
|   |-- SQLite, semantic search, graph, markdown, hierarchical memory
|-- Security
|   |-- encrypted secrets, allowlists, rate limiting, policy engine, audit logs

Documentation

Guide	Use it for
Quick Start	Getting running quickly
User Guide	Daily operation
Developer Guide	Extending Argentum
API Reference	HTTP API and config details
Release Packaging	Binary and installer build details
Security	Security model and reporting
Features	Feature catalog
AI Providers & Extensions	Provider/plugin compatibility and licensing
Update Architecture	Signed updater plan and current limits
FAQ	Curatable in-app help answers
Project Size	Workspace, dependency, artifact, and clean-install measurements

Development

npm install
npm run typecheck
npm run lint
npm test
npm run build

Build Windows release assets from Windows:

npm run package:win

This builds the Windows desktop app installers through Tauri. For the optional portable CLI, run npm run package:win:cli.