Skip to main content

Let language models run code with intelligent action system

Project description

UniShell

AI-powered natural language command execution with cloud gateway integration.

Features

  • Natural Language Commands: Execute system operations using plain English
  • Gateway Integration: Remote command execution through cloud gateway
  • Desktop Agent Mode: Run as background agent for remote management
  • Action Registry: 90+ pre-built actions for system, file, app, and network operations
  • Multi-LLM Support: Works with OpenAI, Azure OpenAI, and local models
  • Safe Execution: Built-in policy engine and rollback capabilities

Installation

pip install unishell-q

Quick Start

Local Interactive Mode

# Basic usage
unishell

# With Azure OpenAI
unishell --model azure/gpt-4o --api_key YOUR_KEY --api_base YOUR_ENDPOINT --api_version 2024-08-01-preview

# With OpenAI
export OPENAI_API_KEY=your-key
unishell --model gpt-4o

# Auto-run mode (skip approval)
unishell -y

Desktop Agent Mode (Gateway Integration)

Run UniShell as a background agent that executes commands from a cloud gateway:

# Start agent with gateway connection
unishell --model azure/gpt-4o --api_key YOUR_KEY --api_base YOUR_ENDPOINT --api_version 2024-08-01-preview --agent

# Configure gateway on first run
Gateway URL: http://192.168.100.227:8000
Device Token: <your-device-token>

# Agent runs in background, CLI accepts commands
> open notepad.exe
> create folder test in C:\Users\Desktop

How Agent Mode Works:

  1. Agent polls gateway for pending commands
  2. Executes commands locally on your desktop
  3. Reports results back to gateway
  4. Cloud UI shows execution status

Gateway Configuration

Agent credentials are stored in ~/.unishell/gateway_config.json:

{
  "gateway_enabled": true,
  "gateway_url": "http://192.168.100.227:8000",
  "device_token": "your-device-token"
}

To reconfigure, delete the file and restart with --agent flag.

Usage Examples

System Operations

> lock my screen
> shutdown system in 60 seconds
> get system info
> check CPU and RAM usage

File Operations

> create a file named test.txt in desktop
> copy file.txt to backup folder
> delete old_file.txt
> rename document.txt to report.txt

Application Management

> open notepad.exe
> install chrome
> force close firefox
> list installed apps

Network Operations

> ping google.com
> flush DNS cache
> check my IP address
> disconnect wifi

Python API

from unishell import unishell

# Configure LLM
unishell.llm.model = "gpt-4o"
unishell.llm.api_key = "your-key"

# Execute commands
unishell.chat("Create a Python script")

# Use gateway orchestrator
from unishell.core.gateway import ExecutionOrchestrator
from unishell.core.intent import LLMIntentClassifier, LLMParameterExtractor
from unishell.core.action_registry import DynamicActionLoader

loader = DynamicActionLoader()
registry = loader.load_all()

classifier = LLMIntentClassifier(llm_client, model_name="gpt-4o")
extractor = LLMParameterExtractor(llm_client, model_name="gpt-4o")

orchestrator = ExecutionOrchestrator(classifier, extractor, registry)
result = orchestrator.execute("open notepad", dry_run=False)

Architecture

Local Mode

User Input → Intent Classifier → Parameter Extractor → Policy Engine → Executor → Result

Gateway Mode

Cloud UI → Gateway (Redis) → Desktop Agent → Local Executor → Gateway → Cloud UI
         ↓                      ↑
    Stores Action          Polls & Reports

Command Line Options

# Model Configuration
--model, -m          Language model to use
--api_key, -ak       API key for LLM
--api_base, -ab      API base URL
--api_version, -av   API version
--temperature, -t    Model temperature

# Agent Mode
--agent, -ag         Start desktop agent mode
--gateway_url, -gu   Gateway URL for agent
--device_token, -dtoken  Device token for authentication

# Execution Options
--auto_run, -y       Automatically run generated code
--verbose, -v        Print detailed logs
--safe_mode          Enable safety mechanisms (off/ask/auto)

# Other Options
--profile, -p        Load configuration profile
--offline, -o        Disable online features
--version            Show version number

Action Registry

UniShell includes 90+ pre-built actions across categories:

  • Application: open, install, uninstall, update, force_close, list
  • File: create, read, write, delete, copy, move, rename, search
  • Folder: create, delete, list
  • System: shutdown, restart, sleep, hibernate, lock, info, resources
  • Network: ping, check_ip, flush_dns, wifi_connect, wifi_disconnect
  • Service: start, stop, restart, enable, disable
  • Process: kill, list
  • Monitoring: cpu_threshold, ram_spike, disk_full, error_logs

Actions are defined in unishell/core/action_registry/actions/.

Gateway Integration

Desktop Registration

  1. Access gateway UI: http://your-gateway:8000
  2. Login with credentials
  3. Register desktop to get device token
  4. Use token with --agent flag

Remote Command Execution

  1. Submit command through gateway UI
  2. Gateway queues action (PENDING)
  3. Desktop agent polls and retrieves action
  4. Agent executes locally (RUNNING)
  5. Agent reports result (COMPLETED/FAILED)
  6. UI displays result

Agent Configuration

First run prompts for configuration:

🤖 Desktop Agent Configuration
==================================================
Gateway URL: http://192.168.100.227:8000
Device Token: <paste-your-token>

Or provide via command line:

unishell --agent --gateway_url http://gateway:8000 --device_token YOUR_TOKEN

Security

  • Policy Engine: Validates permissions before execution
  • Dry Run Mode: Preview actions without executing
  • Path Validation: Prevents directory traversal attacks
  • Safe Mode: Optional code scanning and confirmation
  • Device Authentication: Token-based gateway access

Troubleshooting

Agent Not Connecting

# Check gateway URL is accessible
curl http://192.168.100.227:8000/health

# Verify device token is valid
# Delete config and reconfigure
rm ~/.unishell/gateway_config.json
unishell --agent

Command Execution Fails

# Run with verbose logging
unishell --agent --verbose

# Check action exists
# View available actions in unishell/core/action_registry/actions/

Application Not Found

# Use full executable name
> open notepad.exe  # ✓ Correct
> open notepad      # ✗ May fail

# Check app is installed
> list installed apps

Development

# Clone repository
git clone <repo-url>
cd unishell

# Install in development mode
pip install -e .

# Run tests
pytest tests/

# Add new action
# Create JSON definition in unishell/core/action_registry/actions/
# Implement handler in unishell/core/execution/safe_os_adapter.py

Version

Current Version: 0.5.3

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

unishell_q-0.5.3.tar.gz (273.0 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

unishell_q-0.5.3-py3-none-any.whl (368.9 kB view details)

Uploaded Python 3

File details

Details for the file unishell_q-0.5.3.tar.gz.

File metadata

  • Download URL: unishell_q-0.5.3.tar.gz
  • Upload date:
  • Size: 273.0 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.11.9

File hashes

Hashes for unishell_q-0.5.3.tar.gz
Algorithm Hash digest
SHA256 84ddeb57f57f881107620a10d783d3232df9d715fc1bc70298e38fe0105a602e
MD5 0a0233b123900b3c3f9455d515d0b983
BLAKE2b-256 c7b7be2e205f38cae4630d523ee8b95f148e04b1bece43c73e8cdeebbf01b649

See more details on using hashes here.

File details

Details for the file unishell_q-0.5.3-py3-none-any.whl.

File metadata

  • Download URL: unishell_q-0.5.3-py3-none-any.whl
  • Upload date:
  • Size: 368.9 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.11.9

File hashes

Hashes for unishell_q-0.5.3-py3-none-any.whl
Algorithm Hash digest
SHA256 ecac2adfb661d726e4f3a8354d374254fdf261642d5aedb35991703d2aeb5f86
MD5 e8154cfed74fef4bcfd80b323f94534c
BLAKE2b-256 370292465d5d766a7cf3a2cbbb2cabd281eeccc4c26a43c23e549f0e21ac0287

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page