The Intelligent, Secure, and Elite AI CLI by OpenGit
OpenShell is a professional-grade terminal assistant designed for power users who demand speed, security, and a premium experience. Developed by OpenGit, it bridges the gap between raw shell commands and modern AI intelligence.
With its signature Deep Purple aesthetic and refined interaction model, OpenShell version 2.5.0 introduces native Gemini support, hardened security protocols, and a more obedient agent personality.
- 🧠 Autodidaxis (Self-Learning): First implementation of the persistent memory system. The agent now learns and recalls user preferences.
- ♊ Native Gemini Support: Full integration with Google's Gemini models (including the free API).
- 🛡️ Hardened Security: Multi-layer protection against dangerous commands (
sudo,apt, etc.) even in autonomous mode. - 🎨 Premium UI: Cohesive purple-themed interface with optimized markdown rendering and smooth streaming.
- 🤖 Strict Obedience: Reprogrammed agent personality that follows literal commands and stops hallucinations.
- 🧹 Anti-Watermark Purge: Cleaned up legacy project references to ensure a 100% pure OpenShell experience.
- 🤖 Multi-Provider Intelligence: Support for Gemini, OpenAI, Anthropic, OpenRouter, and Ollama.
- 🧠 Skills System: Expand capabilities by adding Markdown files to
~/.openshell/skills/. - 🔒 Security-First Architecture: Intelligent command flagging and manual confirmation for high-risk actions.
- 📜 Clean Conversation: Auto-cleaning of tool outputs and model-specific history management.
OpenShell is on a journey to become a fully distributed intelligence network. Check out our ROADMAP.md to see how we're implementing:
- Autodidaxis: Continuous self-learning from user interactions.
- Smart Orchestration: Intelligent workflow automation.
- Distributed Connectivity: Multi-node agent coordination.
# Start the premium experience
openshell chat
# Configure your provider (Try the new Gemini option!)
openshell config
# Clean context and start fresh
/clear contextOpenShell 2.1.0 implements a Zero-Trust execution model:
- Sensitive Command Blocking: Patterns like
sudo,rm -rf, and package managers are automatically flagged. - Loop Protection: Capped tool call loops to prevent AI "hallucination spirals".
- User Confirmation: Critical actions always require a manual "Yes" regardless of the autonomy mode.