Install the latest hardened production build:
npm install -g sagar0123/nebula-cli#v5.1.0Tip
🛡️ Security Hardened (v5.1.0)
- Recursive AST Traversal: 100% Block rate against Semantic Escapes.
- Fail Closed: Parser crashes now default to "Block".
- Project Isolation: Memory leakage between directories is physically impossible.
Caution
🚧 UNDER ACTIVE DEVELOPMENT 🚧
This project is currently in an experimental Alpha state. You may encounter:
- Frequent crashes or unexpected behavior.
- Breaking changes between minor versions.
- Hallucinations in AI responses.
Use with caution in production environments.
If you are cloning this repository to build or contribute, ensure you have:
- Node.js: v18.0.0 or higher (Required for ESM support).
- Git: Latest version.
- Ollama: (Optional) For running active local models.
- API Keys: See Configuration section below for Cloud AI access.
"A terminal so smart, it fixes itself before you even notice."
Nebula-CLI is a next-generation terminal agent powered by LLMs (Gemini/Ollama). It doesn't just run commands; it understands your intent, detects failures, and auto-corrects errors in real-time. Whether you're managing Kubernetes clusters or debugging a local Node.js app, Nebula is your silent partner in engineering.
Nebula operates as a resilient, multi-tier intelligent layer between you and the OS kernel, designed for $0.00 cost, zero latency, and maximum privacy.
Your system functions like a spacecraft with redundant engines:
- Primary (Sagar-Private-Space)
- The Brain: Qwen 2.5 7B (iMatrix IQ4_XS) on Hugging Face CPU.
- Role: High-level Tyk architecture planning and complex Kubernetes debugging.
- Secondary (Cloud Burst)
- The Brains: Groq (Llama 70B) & Gemini 2.0.
- Role: Rapid failover if Hugging Face is "Cold Starting" or hitting rate limits.
- Core (Local Fortress)
- The Brain: Qwen 0.5B (Ollama).
- Role: The "Black Box" for basic shell commands. 100% functional offline.
graph TD
classDef primary fill:#ff9900,stroke:#333,stroke-width:2px;
classDef secondary fill:#00ccff,stroke:#333,stroke-width:2px;
classDef core fill:#33cc33,stroke:#333,stroke-width:2px;
classDef security fill:#ff3333,stroke:#333,stroke-width:2px;
User[User Input] -->|Command| Shell[Nebula Shell]
Shell -->|Router| Router{AI Router}
Router -- Training Mode --> Primary[Primary: HF Space]:::primary
Router -- Quick Fix --> Core[Core: Ollama Local]:::core
Router -- Fallback --> Secondary[Secondary: Groq / Gemini]:::secondary
Core -->|Failover| Secondary
Secondary -->|Failover| Primary
Primary & Secondary & Core -->|Response| Scrubber[Secret Scrubber]:::security
Scrubber -->|Safe Answer| Executioner[Dynamic Executioner]
Even with multiple cloud providers, the Secret Masking and Command Validation layers remain local, ensuring secrets are scrubbed before they touch any cloud API.
- Cost: "Pro" grade system for $0.00.
- Latency: "Pre-Warming" and "Local Fallback" ensures zero waiting.
- Privacy: "Red Team" security layer masks secrets locally.
Nebula instantly recognizes what you are working on and adapts its behavior.
- Projects: Helm, RPM, OpenShift, Docker, Terraform, Ansible, Node.js, Generic K8s.
- Environments: Automatically detects Minikube, EKS, GKE, OpenShift, or AKS.
- Result: It generates
aws eks update-kubeconfigfor EKS, butminikube dashboardfor local dev.
Nebula now creates a unique "Memory Vault" for each directory you work in.
- Context Aware: "fix it" knows exactly which error just happened in this specific project.
- Isolation: Your React fixes won't confuse your Python scripts.
- Tiered Lookup:
- Session Cache (0.1ms): Instant recall of recent fixes.
- Project Memory (Local Vector DB): Persistent history for this specific folder.
- Global Wisdom: Universal knowledge shared across all projects.
- Look Before You Leap: Automatically checks Kubernetes connectivity, namespace existence, and missing secrets before running deployment commands.
- Red-Line Warnings: Highlights destructive commands (
rm -rf,kubectl delete,drop table) in RED and demands confirmation. - Safe Execution: Never runs AI commands without your explicit "Yes".
- Smart Timeouts: Knows that
lstakes 1s butdocker buildneeds 10m. - Live Feedback: Real-time progress monitoring (
🔄 156KB | 🟢 Active) instead of frozen screens. - Hung Process Detection: Warns you if a process stops generating output (
🟡 Stalled).
npm install -g nebula-cliConfigure Nebula via .env file or environment variables:
| Variable | Description | Default |
|---|---|---|
GEMINI_API_KEY |
Google Gemini API Key (Required for Cloud AI) | - |
GEMINI_MODEL |
Gemini Model ID | gemini-2.0-flash |
GROQ_API_KEY |
Groq API Key (Alternative Cloud AI) | - |
OLLAMA_MODEL |
Local LLM Model Name | llama3.2 |
Enter the persistent, self-healing shell:
nebula- Ask Anything:
ask "deploy this helm chart to tyk ns" - Auto-Healing: If a command fails, Nebula analyzes the error and suggests a fix.
- Universal: Switch from a Node.js project to a K8s cluster seamlessly.
# Detects project type and suggests next steps
nebula predict
# Analyzes complex failures
nebula ask "why is my pod crashlooping?"Nebula is designed as a copilot, not an autopilot.
- Explicit Consent: Nebula will NEVER execute an AI-suggested command without your explicit "y/N" confirmation.
- Review First: Always read the suggested fix before hitting 'y'.
If you are using the free tier of Gemini or other public LLM providers:
- Data Usage: Your command history and error logs may be processed by human reviewers to improve the model.
- Sensitive Data: DO NOT use Nebula with secrets, API keys, or PII (Personally Identifiable Information) in the terminal output when using public models.
Experimental Technology: Nebula-CLI uses large language models which can be unpredictable.
- Hallucinations: The AI may suggest commands that do not exist or do not solve the problem.
- Liability: You are responsible for the commands executed on your machine. The authors of Nebula-CLI are not liable for any data loss or system damage.
We follow Conventional Commits and strict CI/CD pipelines.
- Fork & Clone
npm installnpm test- Submit a PR with
feat:orfix:messages.
MIT © 2025 Sagar