All Work, via AI
Control your computer
with one line of natural language
AllvIa is a Local OS Super Agent that understands natural language commands and controls apps directly at the OS level. Experience a Rust-native agent that learns your usage patterns via Shadow Mode and suggests high-performance n8n-based automations.
OpenAI
Model
SQLite
Memory
n8n
Tool
지금 웹사이트에서 설명 중인 대상 프로젝트는 allvia-ai/ALLVIA 입니다.
clone, 환경설정, 빌드, 테스트, 릴리스까지 한 흐름으로 확인할 수 있습니다.
curl -L https://github.com/allvia-ai/ALLVIA/archive/refs/heads/main.tar.gz | tar -xz allvia-ai/ALLVIA 저장소 클론
core/.env 파일에 OPENAI_API_KEY 설정
cargo build --manifest-path core/Cargo.toml --release 로 코어 빌드
코어 프로세스 실행: ./core/target/release/local_os_agent
cargo test --manifest-path core/Cargo.toml 로 검증 테스트 실행
./scripts/rebuild_and_deploy.sh 로 앱 번들 재배포
runbook 기준으로 /Applications/Steer OS.app 헬스체크
▌core/src/
├── main.rs # CLI 및 메인 루프
├── intent_router.rs # 자연어 의도 분류
├── plan_builder.rs # 실행 계획 생성
├── execution_controller.rs # 단계 실행
├── verification_engine.rs # 실행 결과 검증
├── policy.rs # 보안 정책 분류
├── security.rs # 셸 명령 위험 분류
├── privacy.rs # PII 마스킹
├── llm_gateway.rs # OpenAI 연동
├── n8n_api.rs # n8n 워크플로우 API
├── visual_driver.rs # UI 자동화 드라이버
└── macos/ # 네이티브 macOS 바인딩
Core Capabilities
99+
Rust Core Modules
5
Security Layers
4
External Integrations
8
CLI Commands
How It Works
An automated pipeline from natural language input to OS control and verification.
Understand
intent_router analyzes natural language, and slot_filler extracts parameters needed for execution.
Execute
plan_builder generates an execution plan, while visual_driver and browser_automation control the OS directly.
Verify
verification_engine validates results, and approval_gate blocks dangerous actions in advance.
Natural Language Control
Execute complex multi-app scenarios with natural language
Handle complex multi-app workflows like "Search stock prices in Safari → Calculate with Calculator → Save in Notes" with a single line of natural language.
-
surf— The core command for controlling the entire OS with natural language -
control— Control app execution, switching, and termination via CLI -
exec— Execute shell commands with security policies applied
$ cargo run --bin local_os_agent
> surf "Open Safari and search for Apple stock price on Google"
Intent: web_search | Slots: {app: Safari, query: Apple stock price}
Plan: Open Safari → Navigate → Search → Extract
[OK] Safari execution completed (applescript.rs)
[OK] Google search completed (browser_automation.rs)
[OK] Result extraction completed (visual_driver.rs)
> control Notes open
[OK] Notes execution completed
> exec "echo hello world"
[Safe] hello world
> exec "rm -rf /"
[BLOCKED] security.rs: Critical — Dangerous command blocked
> exec "sudo apt install"
[BLOCKED] policy.rs: Caution — Write Lock active
> unlock
[OK] Write Lock released (Auto-lock in 60s)
PII Masking (privacy.rs): user@email.com → [EMAIL_MASKED]
Zero Trust Security
Verify LLM output with Zero Trust principles
All AI-generated commands must pass through a 5-layer security engine before execution. Under the "LLM plans, Rust enforces" philosophy, actions require automatic approval, Write Lock release, or explicit user confirmation based on risk levels.
-
policy.rs— 3-level classification: Safe / Caution / Critical -
security.rs— Automatic risk analysis of shell commands -
privacy.rs— Automatic masking of PII like emails and card numbers -
tool_policy.rs— Allow/Denylist management for different tools -
Write Lock— Locked by default, with automatic timed re-locking
Integrations & Automation
Real-time integration with external services
Natively integrate with Gmail, Calendar, and Notion. Remote control your computer with Telegram, and automate with hundreds of services using n8n.
Gmail
Automatically handle email lookup, sending, and classification.
gmail.rs
Calendar
Manage schedules, creation, and reminders.
calendar.rs
Notion
Read/write pages and query databases.
notion.rs
Telegram
Remote control your computer from your phone while away.
telegram.rs
Shadow Mode & Workflow
Learn usage patterns and auto-generate workflows
The Rust-native Collector gathers OS sensor data in the background, and Shadow Mode identifies repeating routines to recommend optimal automation workflows.
-
routine— Automatic analysis of repeating patterns over the last 7 days -
recommend— AI-based automation suggestions -
build_workflow— Natural language to n8n workflow transformation
> routine
Analyzed repeating patterns (Last 7 days):
1. VS Code → Terminal → git commit (3 times daily)
2. Slack → Chrome → Google Docs (5 times daily)
3. Finder → Terminal → npm run dev (4 times daily)
> build_workflow "Send KPI report to Slack every Monday at 9 AM"
Analyzing with LLM...
Generating n8n workflow:
├─ Trigger: Schedule (Monday 09:00 KST)
├─ Action: Google Sheets → Read KPI
├─ Action: Format → Markdown Report
└─ Action: Slack → #team-weekly
[OK] workflow_weekly_kpi.json generated successfully
Desktop & Web UI
From CLI to Desktop App, use it the way you want
Visually manage dashboards, chat, and workflows with our React/Vite-based Web UI, and deploy native desktop apps with Tauri.
Dashboard
Check system status, pattern statistics, and execution logs at a glance.
Chat
Interact with the agent using natural language to give tasks.
Workflows
Manage and edit n8n workflows visually.
Routines
Check detected patterns and automation recommendations.
Audit Log
Track all execution history and security events.
Settings
Configure environment settings, API keys, and security policies.
Frequently Asked Questions
Core questions about installation, security, and features.
Which operating systems are supported?
Based on the current main repository, macOS 12+ (Monterey) is the required/recommended environment. Windows execution is available through scripts/run_core.ps1, and Linux can be compiled via the Rust-native codebase.
Can I control my computer using natural language?
Yes. By entering natural language instructions like "Open Safari and search for weather on Google" with the surf command, the pipeline works automatically: intent_router → slot_filler → plan_builder → execution_controller → verification_engine.
Is an OpenAI API Key required?
Yes. OpenAI API is used for LLM-based behavioral pattern analysis, natural language command interpretation, and workflow generation. You need to set OPENAI_API_KEY in your .env file. GPT-3.5 is sufficient for initial testing.
How is security handled?
We apply Zero Trust principles. LLM output is considered dangerous by default and passes through a 5-layer security model (policy.rs, security.rs, tool_policy.rs, privacy.rs PII masking, send_policy.rs). Write Lock is enabled by default, and Critical-grade actions require per-item user approval.
Which external services are integrated?
Gmail (view/send emails), Google Calendar (check schedules), Notion (read/write pages), and Telegram (remote control) are supported by default. Integrating with the n8n workflow engine allows for automation of hundreds of additional services.
Can teams use it together?
Team deployment is available through Tauri desktop bundles (.app/.exe). It provides n8n workflow sharing and the Self-Healing Guardian path for automatic process restart.
Get Started Today
Cloning, building, and running takes only 5 minutes. Experience the performance of 99+ Rust core modules for yourself.