Skip to content
J.A.R.V.I.S.

Introduction

JARVIS is an always-on autonomous AI daemon built for power users who want an AI that acts, not one that asks permission. It runs as a persistent background service on your machine, maintaining a live model of your world and executing tasks continuously — with or without you at the keyboard.

What Makes JARVIS Different

Section titled “What Makes JARVIS Different”

Most AI tools are reactive. You open them, type a question, read the answer, and close them. JARVIS is something else entirely.

TraitChatbotJARVIS
LifecycleSession-basedAlways running
InitiativeResponds when askedMonitors, schedules, acts proactively
Environment accessSandboxed or noneBrowser, desktop, files, APIs
MemoryCleared each sessionPersistent SQLite knowledge vault
ChannelsOne interfaceDashboard, Telegram, Discord, Voice
Execution modelSingle agentMulti-agent hierarchy with parallel tasks

JARVIS is not a chatbot with tools bolted on. It is an autonomous daemon designed to be dangerously capable by default.

Who JARVIS Is For

Section titled “Who JARVIS Is For”

JARVIS targets power users who:

  • Want an AI that works in the background while they focus on other things
  • Need automation across browser, desktop, files, and external services
  • Communicate through multiple channels (web, Telegram, Discord, voice)
  • Want persistent memory that grows smarter over time
  • Are comfortable granting an AI meaningful autonomy over their machine

If you want an AI that requires constant hand-holding and confirmation dialogs, JARVIS is not the right tool. If you want one that finishes the job, it is.

Key Capabilities

Section titled “Key Capabilities”

Autonomous Browser Control

Section titled “Autonomous Browser Control”

JARVIS auto-detects and launches Chrome or Chromium, attaches via the Chrome DevTools Protocol, and can navigate, click, type, extract content, and take screenshots — all without you touching the browser.

Native Desktop Control

Section titled “Native Desktop Control”

On Windows (including WSL), JARVIS launches a C# sidecar that uses the FlaUI/UIA automation framework to control any desktop application: find windows, click buttons, read text, type into fields, and scroll.

Voice Interface

Section titled “Voice Interface”

Speak to JARVIS using a wake word (“Hey JARVIS”), microphone button, or push-to-talk. Responses are streamed back as speech using edge-tts-universal — no API key required for TTS. Speech-to-text supports OpenAI Whisper, Groq, or a local model.

Multi-Agent System

Section titled “Multi-Agent System”

JARVIS orchestrates a hierarchy of 11 specialist sub-agents. A primary agent delegates tasks to specialists (researcher, coder, writer, analyst, and others) either synchronously for single tasks or asynchronously in parallel for complex workflows.

Persistent Memory

Section titled “Persistent Memory”

Every conversation contributes to a SQLite knowledge vault. JARVIS automatically extracts facts, preferences, events, and relationships from responses and injects relevant knowledge into future conversations.

Proactive Behavior

Section titled “Proactive Behavior”

JARVIS does not wait to be asked. It monitors Gmail and Google Calendar, executes scheduled commitments, queues research tasks, and sends you notifications through D-Bus (Linux), Telegram, or Discord.

Communication Channels

Section titled “Communication Channels”

Interact through the web dashboard, Telegram bot, Discord bot, or voice — all unified in a single conversation history.

Architecture Overview

Section titled “Architecture Overview”
┌─────────────────────────────────────────────────────────┐
│                    JARVIS Daemon                         │
│                                                         │
│  ┌─────────────┐  ┌──────────────┐  ┌───────────────┐  │
│  │  Agent       │  │  Memory      │  │  Proactive    │  │
│  │  Orchestrator│  │  Vault       │  │  Engine       │  │
│  │  (+ sub-     │  │  (SQLite)    │  │  (observers,  │  │
│  │   agents)    │  │              │  │   scheduler)  │  │
│  └──────┬───────┘  └──────────────┘  └───────────────┘  │
│         │                                                │
│  ┌──────▼───────────────────────────────────────────┐   │
│  │              Tool Layer                           │   │
│  │  Browser  Desktop  Files  Search  APIs  Shell    │   │
│  └──────────────────────────────────────────────────┘   │
└─────────────────────────┬───────────────────────────────┘
                          │ WebSocket + REST
        ┌─────────────────┼─────────────────┐
        ▼                 ▼                 ▼
   Web Dashboard      Telegram Bot      Discord Bot
   (localhost:3142)   (@yourbot)        (#channel)
                          │
                     Voice (mic/TTS)

The daemon exposes a WebSocket server at localhost:3142 by default. The web dashboard connects to this endpoint for real-time streaming. Telegram and Discord adapters connect the same agent loop to external messaging platforms. All channels share a unified conversation history.

Technology Stack

Section titled “Technology Stack”
  • Runtime: Bun — fast TypeScript runtime, SQLite built in
  • Language: TypeScript (ESM modules)
  • LLM: Anthropic Claude (primary), OpenAI GPT-4 (fallback), Ollama (local)
  • Database: SQLite via bun:sqlite
  • Browser automation: Chrome DevTools Protocol (CDP)
  • Desktop automation: C# FlaUI sidecar over TCP
  • TTS: edge-tts-universal (free, no API key)
  • STT: OpenAI Whisper / Groq / local Whisper
  • Wake word: openwakeword-wasm-browser (ONNX models, runs in-browser)
  • Frontend: React

Next Steps

Section titled “Next Steps”