Krave CLI

A terminal-native AI coding assistant.
Engineered for software professionals who work in the shell. Krave connects you to proprietary AI models purpose-built for software engineering — delivering real-time code generation, debugging, file management, and system automation without leaving your terminal.

Overview

Krave is not another web-based chat interface. It is a fully native terminal application that integrates AI-assisted development into your existing workflow. By operating directly in the shell, Krave provides:

• Real system access — executes bash commands, reads and writes files, runs git operations and build tools on your actual machine
• Autonomous subagents — spawns independent AI workers for parallel task execution
• Persistent context — maintains full session history with automatic save and resume
• Zero overhead — ships as a single pre-built file with no runtime dependencies

Features

Dual-Model Architecture

Krave provides two proprietary AI models, selectable mid-session:

| Model | Profile | Optimal Use | |-------|---------|-------------| | Krave 4.1 | Fast reasoning, low latency | Iterative development, quick edits, file operations, debugging cycles | | Krave 4.2 | Deep analytical capability | Complex refactoring, architectural decisions, multi-file changes, code review |

Switch between models at any time with /model — conversation context is preserved across switches.

Tool System

The AI can invoke five distinct tools in response to your requests:

| Tool | Function | |------|----------| | bash | Executes shell commands directly on your machine with a 30-second timeout | | writefile | Creates or modifies files with inline diff display | | spawn | Delegates tasks to subagents that work independently and return structured results | | webSearch | Queries the web for real-time information — documentation, packages, current events | | webFetch | Retrieves full-page content from a specified URL |

Session Management

• Conversations are saved automatically to the local chats/ directory
• Resume any previous session with /load
• Export session data with /export
• Track token consumption and API usage with /tokens and /stats

Installation

npm install -g krave-cli

Requirements

• Node.js >= 18.0.0
• A terminal emulator with Unicode and ANSI color support

Platform Support

| Platform | Status | |----------|--------| | Linux | Full support | | macOS | Full support | | Windows (WSL) | Full support | | Windows (CMD / PowerShell) | Supported | | Termux (Android) | Supported |

Usage

Starting Krave

krave

The terminal UI initializes with session information and awaits your input.

Command Reference

| Command | Description | |---------|-------------| | /help | Display the command reference | | /new | Start a new session (current session auto-saves) | | /load [n] | List saved sessions or load one by index | | /save | Persist the current session manually | | /model | Switch between Krave 4.1 and Krave 4.2 | | /context | Display current session configuration | | /stats | Show session statistics (message count, duration, token usage) | | /history | Review all messages in the current session | | /tokens | Display detailed token consumption | | /export | Export the current session as JSON | | /clear | Clear chat history without saving | | /version | Show version and model information | | /exit or /quit | Save the current session and terminate |

Session Example

  ┌── KRAVE v2.0.1 ──────────────────┐
  │ Krave AI · by Asta · Krave 4.2   │
  │ chats: 3 · 14 msgs left · 0 msgs │
  │ 5/8/2026, 10:30:00 AM             │
  └───────────────────────────────────┘
  /help for commands

  ├ you  refactor the authentication middleware to use async/await

  ├ krave yeah, let me look at the current implementation first

  ├ bash  cat src/middleware/auth.js
  │ ╰─ [file contents]
  │

  ├ krave swapping the callback chain to async/await. here's the refactor

  ├ write  src/middleware/auth.js
  │ ╰─ [diff displayed]
  │

Architecture

Krave operates on a tool-calling loop:

1. The user submits a request at the prompt
2. The AI model generates a response, optionally invoking one or more tools
3. Tools execute in real-time on the local machine (bash, file I/O, web search)
4. Results are returned to the model for further processing
5. Steps 2–4 repeat until the task is complete or the step limit is reached
6. Sessions are persisted to disk for future resumption

All communication between the CLI and the AI infrastructure occurs over HTTPS. No third-party APIs are involved — Krave uses proprietary hosted models.

Troubleshooting

krave: command not found

Ensure the global npm binary directory is in your PATH:

npm config get prefix

Add the following to your shell configuration (~/.bashrc, ~/.zshrc, or equivalent):

export PATH="$(npm config get prefix)/bin:$PATH"

Reload your shell or run source ~/.zshrc.

Terminal Display Issues

Krave uses box-drawing characters (┌──│└┘) and ANSI color codes. Verify that your terminal emulator supports:

• Unicode (UTF-8) rendering
• 8-bit / 24-bit ANSI color
• Terminal width of at least 60 columns

Most modern terminals — including GNOME Terminal, iTerm2, Windows Terminal, and Termux — meet these requirements.

License

MIT — Built by Asta