Open Source Alpha

The assistant that actually works for you

EmiOS is a local-first personal AI assistant that manages your day, controls your home, browses the web, and learns who you are — all running on your machine, under your control.

Get Started How It Works
65+
Specialized Agents
40+
Built-in Tools
100%
Local-First

What Emi Can Do

Not a chatbot. A system of agents that take action on your behalf.

📋

Day Flow Orchestrator

Plans your day, tracks tasks across plans, dispatches actions at the right time, and adapts when things change. Strategic planning meets real-time execution.

📧

Email, Calendar & Tasks

Reads your email, triages by importance, manages calendar events, creates reminders. Sends emails in your voice with your signature.

🌐

Browser Automation

Playwright-powered real browser control. Orders food, researches the web, fills forms, captures screenshots — with visual understanding of what's on screen.

🏠

Smart Home Control

Nest thermostat, smart lights, Ring cameras. Sets temperatures based on your schedule, adjusts lighting, checks who's at the door.

🧠

Knowledge Graph Memory

Builds a persistent graph of your life — people, places, preferences, relationships. Entity cards with beliefs, evidence tracking, and automatic deduplication.

💬

Multi-Room Conversations

Slack, Telegram, SMS, web UI — each with its own identity, safety policy, and conversation scope. Talk to Emi from anywhere.

Built Different

Not a wrapper around an API. A complete agent operating system.

Multi-Agent Orchestration

  • Declarative agents: YAML config + Jinja2 prompts + Pydantic forms
  • Blackboard scope stack for agent-to-agent communication
  • Manager pattern: delegator routes, planner decides, tools execute
  • Critic and summary agents keep long tasks on track

Day Flow Lifecycle

  • Items flow: new → triaged → actionable → dispatched → completed
  • Strategic planner creates multi-step plans with dependencies
  • Action selector dispatches one task per tick
  • State mover promotes waiting items when conditions are met

Local-First Stack

  • Flask + SQLite + ChromaDB — runs on your laptop
  • All data stays on your machine
  • Pluggable LLM providers: OpenAI, Gemini, Anthropic, local
  • MCP protocol for extensible tool integration

Scope & Safety

  • Inherited scope chain: Room → Manager → Agent → Tool
  • Scope can only narrow, never expand
  • Tool approval gates for dangerous operations
  • Per-room identity, permissions, and conversation policy

Design Principles

The philosophy behind EmiOS.

Deterministic Over Probabilistic

Control nodes make routing decisions. LLMs make content decisions. The architecture doesn't depend on the LLM being reliable.

Agents Are Hands, Not Brains

Action agents pick from what's given to them. The thinking happens in planners. The filtering happens in deterministic code.

Scope Can Only Narrow

A child process never has more permissions than its parent. Tools, resources, and actions are constrained at every level.

Local-First, Always

Your data lives on your machine. Your conversations stay private. Cloud APIs are used for inference, not storage.

Fail Loud, Recover Gracefully

No silent fallbacks. Errors surface immediately. But the system degrades gracefully — a failed tool doesn't crash the day.

Observe, Then Act

Snapshot first, then visual scout, then coordinates. Understand the state before taking action.

Ready to meet Emi?

EmiOS is in alpha. It's rough around the edges. But it's real, it runs, and it's yours.

View on GitHub