Local AI System — Three autonomous agents with unified memory
Project description
LAIS — Local AI System
Three autonomous agents. One shared brain. Zero cloud dependency.
What is LAIS?
LAIS is a multi-agent AI operating system that runs three autonomous agents simultaneously on your local machine:
| Agent | Interface | Purpose |
|---|---|---|
| JARVIS (Mark XXXIX) | Voice + Vision | Real-time voice AI via Gemini Live, screen/webcam analysis, desktop control, security grid |
| AI Engine | Desktop GUI | Plugin orchestrator with 40+ hot-loaded plugins, RAG pipeline, local LLM inference, multi-agent routing |
| OpenCode | CLI Terminal | 30+ coding skills: TDD, refactoring, code review, research, debugging |
All three agents share a unified memory layer — JARVIS remembers what OpenCode did, AI Engine orchestrates complex multi-step workflows, and OpenCode handles precise code operations. They communicate through the CoComm cross-agent protocol with A2A server, WebSocket messaging, shared memory, trust scoring, and consensus.
What Makes LAIS Unique
While projects like OpenClaw (374K stars), Hermes Agent (140K stars), and AutoGPT (150K stars) have pioneered the AI agent space, LAIS occupies a distinct architectural niche:
1. Triple-Agent Architecture
No other open-source system runs voice AI + GUI orchestrator + CLI coder simultaneously with shared memory. OpenClaw is messaging-first. Hermes is CLI-only. LAIS has all three modalities in one system.
2. 4-Layer Memory Architecture (v3.0)
Hot (100%) → Full context, current session
Warm (60%) → Summarized recent interactions
Cold (20%) → Metadata only, full compression
Crystallized (90%) → Key learnings, permanent storage
Neither OpenClaw nor Hermes tiers memory by compression level with graduated retention.
3. Token Optimization Pipeline (v1.0.0)
Four compression engines in a single pipeline with per-agent USD budgeting:
- claw-compactor — 14-stage content-type-aware compression
- LLMLingua — Microsoft 20x semantic compression
- tokenpruner — 40-60% dedup compression (COMPOSITE strategy)
- shekel — Per-agent USD budget enforcement (warn at 80%, stop at 100%)
- ResponseCache — TTL-based response deduplication
4. 9-Agent Security Grid
Dedicated security sub-agents built into JARVIS:
network_shield · code_sentry · file_watchdog · input_sanitizer · auth_gate · anomaly_detector · crypto_guard · audit_logger · decoy_engine
5. CoComm Cross-Agent Protocol (16 Modules)
| Module | Purpose | Module | Purpose |
|---|---|---|---|
| A2A Server | Agent-to-agent task delegation | Session Log | Active session tracking |
| WebSocket | Real-time messaging | Shared Memory | Cross-agent memory store |
| MCP Bridge | Model Context Protocol | Config | Configuration management |
| Vault Sync | Knowledge base sync | Roles | Agent role definitions |
| Trigger | Event-driven triggers | Handoff | Task handoff protocols |
| Async Agent | Async execution | Goal Planner | Multi-agent planning |
| Consensus | Decision consensus | Graph Evolution | Dynamic knowledge graph |
| Trust | Trust scoring/validation | Memory Sync | Memory synchronization |
6. Knowledge Vault Integration
LAIS uses an Obsidian vault as its source of truth — all shared memory, protocols, agent registries, and crystallized learnings live in a structured, queryable knowledge base. Bi-directional sync means the vault updates from agent activity and agents query the vault for context.
7. Windows Native
While OpenClaw and Hermes target Linux/Mac, LAIS is born on Windows 11 with PowerShell-native automation, Windows Task Scheduler integration, and native Windows desktop control.
System Architecture
┌──────────────────────────────────────────────────────────────────┐
│ LAIS SYSTEM │
├──────────────────────────────────────────────────────────────────┤
│ │
│ ┌──────────────────┐ ┌──────────────────┐ ┌────────────────┐ │
│ │ JARVIS (Voice) │ │ AI Engine (GUI)│ │ OpenCode (CLI) │ │
│ │ Mark XXXIX │ │ 40+ Plugins │ │ 30+ Skills │ │
│ │ Gemini Live │ │ RAG Pipeline │ │ Code Ops │ │
│ │ Screen Vision │ │ Local LLM │ │ Refactoring │ │
│ │ Desktop Ctrl │ │ Orchestrator │ │ TDD/Review │ │
│ │ Security Grid │ │ Self-Improve │ │ Research │ │
│ └────────┬─────────┘ └────────┬─────────┘ └───────┬────────┘ │
│ │ │ │ │
│ └─────────────────────┼─────────────────────┘ │
│ │ │
│ ┌────────────┴────────────┐ │
│ │ UNIFIED LAYER │ │
│ │ Memory · Routing · Sync │ │
│ │ Token Optimization │ │
│ │ 60 Integration Modules │ │
│ └────────────┬────────────┘ │
│ │ │
│ ┌────────────┴────────────┐ │
│ │ KNOWLEDGE │ │
│ │ Obsidian Vault Sync │ │
│ │ Crystallized Memory │ │
│ │ RAG · SQLite FTS5 │ │
│ └─────────────────────────┘ │
└──────────────────────────────────────────────────────────────────┘
Quick Start
Prerequisites
- Windows 10/11 (primary target; Linux/Mac via WSL2)
- Python 3.11+
- ~2GB free disk space (source code only; models require additional)
One-Line Install
powershell -c "irm https://raw.githubusercontent.com/StefSNS/LAIS/main/install.ps1 | iex"
Or via pip
pip install lais-ai
lais install
Manual Install
git clone https://github.com/StefSNS/LAIS.git
cd LAIS
python install.py
Start All Agents
.\launch\start_all.ps1
# Or individually:
python models\Mark-XXXIX\main.py # JARVIS voice AI
python models\ai_engine\main.py # AI Engine GUI
lais_opencode.py # OpenCode launcher
Use Cases
| Scenario | How LAIS Handles It |
|---|---|
| "Research quantum computing and build a demo" | AI Engine RAG-searches knowledge base → drafts report → delegates code to OpenCode |
| "What's on my screen? Open that file." | JARVIS captures screen via Gemini Vision → identifies file → opens it |
| "Review and refactor this module" | OpenCode runs code review skill → applies refactoring → JARVIS announces completion |
| "Set a reminder for my 3pm meeting" | JARVIS captures voice → schedules Windows task → confirmation spoken |
| "Find that email about the API key from last week" | AI Engine semantic search across memory + email plugin → returns result |
| "Monitor my system health" | JARVIS security grid runs diagnostics → AI Engine logs to vault → OpenCode creates report |
Comparison with Other AI Systems
| Feature | LAIS | OpenClaw | Hermes Agent | AutoGPT |
|---|---|---|---|---|
| Voice AI | Native (Gemini Live) | No | No | No |
| Desktop GUI | CustomTkinter | Web UI only | No | No |
| CLI Agent | OpenCode skills | Built-in | Built-in | Built-in |
| Triple Interface | Voice + GUI + CLI | Messaging only | CLI only | CLI only |
| Tiered Memory | 4-layer (v3.0) | Flat persistence | Persistent files | Flat |
| Token Optimization | 4-engine pipeline | None | None | None |
| Security Grid | 9 dedicated agents | Prompt guard | None | None |
| Cross-Agent Protocol | 16-module CoComm | None | None | None |
| Knowledge Vault | Obsidian sync | None | None | None |
| Per-Agent Budgeting | shekel enforcement | None | None | None |
| Self-Improving Skills | Manual | Auto (Hermes) | Auto | Limited |
| Messaging Platforms | None | 14+ providers | 14+ providers | None |
| Windows Native | Yes | Secondary | WSL2 | Secondary |
| GitHub Stars | New | 374K | 140K | 150K |
Directory Structure
LAIS/
├── install.py # Bootstrap installer
├── lais_opencode.py # OpenCode launcher
├── auto_loader.py # Session start protocol
├── README.md # This file
├── LICENSE
├── models/
│ ├── Mark-XXXIX/ # JARVIS voice AI
│ │ ├── main.py # Entry point (PyQt6 + Gemini Live)
│ │ ├── ui.py # Desktop UI
│ │ ├── actions/ # 17 action modules
│ │ │ ├── browser_control.py
│ │ │ ├── desktop.py
│ │ │ ├── screen_processor.py
│ │ │ ├── send_message.py
│ │ │ ├── web_search.py
│ │ │ └── ... (17 total)
│ │ ├── agency/ # Security agency (9 agents)
│ │ ├── core/ # System prompt
│ │ ├── memory/ # Memory manager
│ │ └── config/ # API key template
│ └── ai_engine/ # AI Engine orchestrator
│ ├── main.py # CustomTkinter GUI
│ ├── llm_engine.py # LLM inference gateway
│ ├── plugin_manager.py # Hot-loads 40+ plugins
│ ├── plugins/ # Plugin modules
│ ├── unified_layer/ # 60 integration modules
│ │ ├── token_optimizer.py
│ │ ├── memory_sync.py
│ │ ├── skill_engine.py
│ │ ├── rag_pipeline.py
│ │ ├── orchestrator.py
│ │ ├── a2a_server.py
│ │ └── ... (60 total)
│ ├── knowledge/ # RAG, memory, skills
│ ├── mcp_servers/ # MCP servers
│ └── local_llm/ # Local LLM scripts
├── config/
│ └── system.json # Shared configuration
├── launch/
│ └── start_all.ps1 # Launch all agents
├── addons/
│ └── token-optimizer/ # Token optimization v1.0.0
└── integrations/ # External tool configs
Token Optimization Layer
LLM Call → CompressionPipeline → claw-compactor + tokenpruner → Compressed Prompt → LLM
↓
Shell Command → ShellCompressor → sqz compressor → Compressed Output → LLM
↓
Any LLM Call → TokenBudget → shekel cost tracking → Block if over budget
↓
All operations → Token Log → Usage stats & reporting
from unified_layer.token_optimizer import get_token_optimizer
opt = get_token_optimizer("jarvis")
opt.get_report() # Full token usage + savings report
Environment variables: LAIS_TOKEN_OPTIMIZATION=1, LAIS_SQZ_ENABLED=1, LAIS_BUDGET_ENABLED=1
Community & Roadmap
v2.0 — Current Release
- Three-agent architecture (JARVIS + AI Engine + OpenCode)
- 60 unified_layer integration modules
- Token Optimization v1.0.0 pipeline
- Memory Architecture v3.0 (4-layer)
- CoComm 16-module cross-agent protocol
- JARVIS Mark XXXIX with Gemini Live voice
- 9-agent security grid
- Obsidian vault sync
v2.1 — Current Release
- Messaging gateway (Telegram, Discord, WhatsApp) —
pip install lais-ai[messaging] - One-line pip install (
pip install lais-ai) - Docker deployment
- Linux native support
- Self-improving skills engine
- Model marketplace
v3.0 — Vision
- Agent social network (MoltBook-style)
- Community skill marketplace
- Web UI dashboard
- Mobile companion app
- Plugin SDK for third-party developers
License
CC BY-NC 4.0 — Personal and non-commercial use only. Commercial licenses available upon request.
Security
This system has full access to your computer. Review SECURITY.md before deployment. JARVIS includes a 9-agent security grid for defense-in-depth, but ultimate responsibility rests with the user.
Built with Python, CustomTkinter, PyQt6, Gemini Live API, and way too much coffee.
LAIS — GitHub
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file lais_ai-2.0.0.tar.gz.
File metadata
- Download URL: lais_ai-2.0.0.tar.gz
- Upload date:
- Size: 215.9 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.13.7
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
2891533f392cbe687f3f2a4e934d254eec93c7524c41fd00a6bfceadf59d487e
|
|
| MD5 |
7481b0e28c75f2f67789407c2d69e860
|
|
| BLAKE2b-256 |
55f069b758cb444b67d447e3af9794d1d2258a4d03a872bb564870b8d6d0e609
|
File details
Details for the file lais_ai-2.0.0-py3-none-any.whl.
File metadata
- Download URL: lais_ai-2.0.0-py3-none-any.whl
- Upload date:
- Size: 253.2 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.13.7
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
62bfbc59aa82ecc951060b9dad2e587fb9b49bbc19c46f9c65993d5403b01dfc
|
|
| MD5 |
dd5c728c83fb1f6649cc8da63669dd4f
|
|
| BLAKE2b-256 |
fce2f7b09dc8bed1bfcb3b7d9329d72d7a136431c1187919abdd1a0a9a017484
|