TigrimOS v1.4.0

A self-hosted AI workspace with chat, code execution, parallel multi-agent orchestration, cross-machine remote agents, Auto (AI create architecture), live agent diagram, async parallel sub-agents, and a skill marketplace. Runs on macOS and Windows. Everything executes inside a secure Ubuntu sandbox — no Docker required.

AI-generated code and shell commands cannot escape the sandbox or touch your files without permission. Mix different AI providers in the same agent team — OpenAI-compatible APIs, Claude Code CLI, and Codex CLI. Delegate tasks to remote TigrimOS instances running on other machines — the orchestrator chooses the right agent based on persona and responsibility. Auto mode lets the AI analyze your prompt, design a custom multi-agent architecture (YAML), and boot all agents automatically — no manual configuration needed. Live Agent Diagram shows real-time orchestrator/worker graphs with status badges, tool tracking, and edge states. Agents work in true parallel using async task dispatching with improved P2P swarm governance. Connect external MCP servers to extend the AI's toolbox. Built with 16 built-in tools and designed for long-running sessions with smart context compression and checkpoint recovery.

What's New in v1.4.0

• Host Folders — browse and edit files on your host machine via VM shared folders directly from the Files page. UTM/VirtFS, VirtualBox, and VMware shared folders are auto-detected with proper labels and permissions. The AI agent is aware of all connected host folders and can read/write them.
• Per-message feedback — thumbs up/down and comment buttons on each assistant message in chat, feeding into the skill auto-update synthesizer so liked answers reinforce skills and disliked ones get corrected.
• VM shared folder auto-detection — parses /proc/mounts to identify 9p, virtiofs, vboxsf, and VMware FUSE mounts. Uses the mount tag (UTM share name) as the label instead of generic shared-0 names. Host-origin folders are tagged with a green "Host" badge.
• Files page tabs — Sandbox and Host Folders tabs in the Files page. Local Files removed from sidebar to reduce confusion.
• Agent host folder awareness — all enabled host folder mounts with paths and permissions are injected into the AI agent's system prompt, so the agent can read from and write to shared folders.
• Cross-browser local file access — rewrote Local Files from Chrome-only File System Access API to server-side API, working in all browsers.
• About section in Settings — shows version, build number, and copyright.

Security first: Everything runs inside a real Ubuntu sandbox. Your host file system is completely invisible to the AI unless you explicitly share a folder.

Screenshots

AI Chat with tool-calling — generates React/Recharts visualizations rendered in the output panel.

Visual Agent Editor — drag-and-drop multi-agent design with mesh networking and YAML export.

Minecraft Task Monitor — live pixel-art agents with speech bubbles, walking animations, and inter-agent interactions.

Live Agent Diagram — real-time orchestrator/worker graph with status badges, tool call tracking, edge states, Bus activity bar, and chat log panel.

Benchmark

FrontierScience-Olympiad accuracy — Minimax 2.7 as a single agent scores 62%. With TigrimOS multi-agent orchestration, the same model reaches 75%, surpassing Claude Opus 4.5 (71.4%) and Grok 4 (66.2%), and approaching Gemini 3 Pro (76.1%) and GPT-5.2 (77.1%).

Downloads

Download from the latest release:

Platform	Download	Sandbox Technology
macOS — Apple Silicon (M1/M2/M3/M4)	TigrimOS-v1.4.0-macOS-AppleSilicon.zip	Apple Virtualization.framework
macOS — Apple Silicon (macOS 26 Tahoe)	TigrimOS-v1.4.0-macOS-Tahoe-AppleSilicon.zip	Apple Virtualization.framework
macOS — Intel	TigrimOS-v1.4.0-macOS-Intel.zip	Apple Virtualization.framework
Windows 10/11	TigrimOS-v1.4.0-Windows.zip	WSL2 (Windows Subsystem for Linux)

Requirements

macOS

• macOS 13.0 (Ventura) or later
• Homebrew with qemu (Intel only: brew install qemu)
• 4 GB RAM available for the VM
• ~5 GB disk space (Ubuntu image + TigrimOS)

Windows

• Windows 10 version 2004+ or Windows 11
• WSL2 support (enabled automatically by the installer)
• 4 GB RAM available for the WSL2 instance
• ~5 GB disk space (Ubuntu + TigrimOS)

Installation

macOS

1. Install Homebrew if you don't have it: bash /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
2. Intel Macs only — install qemu (needed to convert the disk image): bash brew install qemu
3. Download the release zip for your Mac:
4. Apple Silicon (M1/M2/M3/M4): TigrimOS-v1.4.0-macOS-AppleSilicon.zip
5. Intel: TigrimOS-v1.4.0-macOS-Intel.zip
6. Unzip — you get TigrimOS.app (or TigrimOS_i.app) and tiger_cowork/ folder
7. Keep both in the same directory (the app needs tiger_cowork/ next to it)
8. Double-click the .app to launch
9. First launch: if macOS blocks it, right-click → Open, or go to System Settings → Privacy & Security → Open Anyway
10. Wait ~5-10 minutes for the Ubuntu sandbox to provision on first run

That's it. Subsequent launches start in under a minute.

Windows — Installer

1. Download and unzip TigrimOS-v1.4.0-Windows.zip
2. Double-click TigrimOSInstaller.bat
3. The graphical installer will guide you through:
4. Enabling WSL2 (may require a one-time restart)
5. Installing Ubuntu 22.04 as a dedicated "TigrimOS" WSL2 distribution
6. Installing Node.js 20 + Python 3 inside the sandbox
7. Optionally connecting a shared folder (can also be done later from the app)
8. Cloning, building, and starting TigrimOS
9. TigrimOS opens as a standalone desktop window (Edge app mode — no browser tabs or address bar)
10. A desktop shortcut TigrimOS is created automatically

After installation, use TigrimOSStart.bat (or the desktop shortcut) to launch and TigrimOSStop.bat to stop.

Install from Git (Alternative)

If you prefer to install from source instead of downloading the release zip:

macOS:

git clone https://github.com/Sompote/TigrimOS.git
cd TigrimOS
xattr -cr TigrimOS.app        # Apple Silicon (M1/M2/M3/M4)
open TigrimOS.app
# or
xattr -cr TigrimOS_i.app      # Intel
open TigrimOS_i.app

Windows:

git clone https://github.com/Sompote/TigrimOS.git
cd TigrimOS
powershell -ExecutionPolicy Bypass -File install_windows.ps1

Note (macOS): Run the app from inside the cloned folder — tiger_cowork/ must be next to the .app for the VM to find it.

Quick Start

1. Launch TigrimOS
2. macOS: Open the app — the setup wizard runs on first launch
3. Windows: Double-click TigrimOSStart.bat or the desktop shortcut — opens as a standalone app window
4. Wait for the Ubuntu sandbox to provision (~5-10 minutes on first launch)
5. Open Settings → enter your API Key, API URL, and Model
6. Click Test Connection to verify
7. Start chatting — the AI can search the web, run code, generate charts, and more

Subsequent launches start in under a minute (no re-download).

Connect a Local LLM (Ollama, llama.cpp, LM Studio)

TigrimOS can use AI models running on your host machine — no cloud API key needed.

Step 1: Start your local model server on 0.0.0.0

The server must listen on 0.0.0.0 (all interfaces), not 127.0.0.1. The sandbox connects through a network bridge, so localhost-only servers are unreachable.

llama.cpp / llama-server:

llama-server -hf LiquidAI/LFM2.5-1.2B-Instruct-GGUF -c 4096 --port 8080 --host 0.0.0.0

Ollama:

OLLAMA_HOST=0.0.0.0 ollama serve

LM Studio: In LM Studio settings → Server → set host to 0.0.0.0, then start the server.

Step 2: Configure TigrimOS

In the TigrimOS web UI, go to Settings → AI Provider:

Field	llama.cpp	Ollama	LM Studio
Provider	OpenAI-Compatible (Local)	Ollama (Local)	LM Studio (Local)
API URL	http://host.local:8080/v1	http://host.local:11434/v1	http://host.local:1234/v1
Model	Your model name (e.g. LiquidAI/LFM2.5-1.2B-Instruct-GGUF)	llama3.2, mistral, etc.	local-model
API Key	local (any text)	local (any text)	local (any text)

macOS: host.local is a special hostname inside the VM that routes to your Mac. It's set up automatically during provisioning.

Windows: host.local resolves to your Windows host via WSL2 networking. If it doesn't work, use your PC's local IP address (e.g. 192.168.1.x).

Step 3: Test Connection

Click Test Connection in Settings. If it succeeds, you're ready to chat.

Troubleshooting Local LLM

Problem	Solution
"fetch failed"	Make sure the server is running with --host 0.0.0.0
"Connection error"	Check the port number matches your server
"host.local not found"	macOS: Click Reset VM in toolbar → restart the app. Windows: Use your PC's IP instead
Server works in browser but not in TigrimOS	Your server is on 127.0.0.1 — restart with 0.0.0.0

Key Features

• AI Chat with 16 Built-in Tools — web search, Python, React, shell, files, skills, sub-agents
• Mix Any Model per Agent — assign different AI providers per agent (API, Claude Code CLI, Codex CLI)
• Parallel Multi-Agent System — 7 orchestration topologies (hierarchical, mesh, hybrid, P2P, P2P+orchestrator, pipeline, broadcast), 4 communication protocols, P2P swarm governance with blackboard bidding
• Swarm Communication Protocols — TCP (private 1-on-1 channels), Bus (broadcast to all), Blackboard (P2P auction: propose → bid → award → execute), Mesh (any agent can talk to any other)
• Remote Agents — delegate tasks to TigrimOS instances on other machines over the network; orchestrator auto-selects agents by persona and responsibility; fully peer-to-peer (any machine can be orchestrator or worker)
• Built-in Terminal — full xterm.js terminal with root access to the Ubuntu sandbox (install packages, manage services, run CLI tools)
• Minecraft Task Monitor — live pixel-art characters with speech bubbles showing agent activity and remote progress
• Long-Running Session Stability — sliding window compression, smart tool result handling, checkpoint recovery
• MCP Integration — connect any Model Context Protocol server (Stdio, SSE, StreamableHTTP)
• Output Panel — renders React components, charts, HTML, PDF, Word, Excel, images, and Markdown
• Skills & ClawHub — install AI skills from the marketplace or build your own
• Projects — dedicated workspaces with memory, skill selection, and file browser
• Cross-Platform — native macOS app + Windows WSL2 installer

Sandbox Terminal

TigrimOS includes a built-in terminal that gives you root access to the Ubuntu sandbox. It runs a real PTY with full color, tab completion, and cursor support via xterm.js.

How to Open the Terminal

Platform	How
Web UI (macOS & Windows)	Click Terminal in the sidebar navigation
macOS native app	Click the Terminal button in the top toolbar (available when VM is running) — opens macOS Terminal.app with SSH into the VM. Password: tigris
Windows	Double-click TigrimOSTerminal.bat — opens a command prompt directly into the WSL2 Ubuntu sandbox (no password needed)

Use the terminal to install additional tools, manage services, or debug the sandbox environment.

First-Time Setup: Claude Code CLI

1. Open the Terminal (sidebar → Terminal)
2. Install and login: bash npm i -g @anthropic-ai/claude-code ln -sf /root/.local/bin/claude /usr/local/bin/claude claude login A URL will appear — open it in your browser and authorize. That's it.

Or use an API key instead: bash echo 'export ANTHROPIC_API_KEY=sk-ant-...' >> /root/.bashrc && source /root/.bashrc

First-Time Setup: Codex CLI

1. Open the Terminal (sidebar → Terminal)
2. Install and login: bash npm i -g @openai/codex codex login --device-auth A