AI-powered CLI: explain code, generate shell commands, use any LLM provider.

Built in Rust. Works on macOS, Linux, and Windows.

$ niko cmd "find all files larger than 100MB"
find . -type f -size +100M
Copied to clipboard

$ cat main.rs | niko explain
📖 42 lines analyzed — completed in 2.1s
## Overview
...

• Three Modes — cmd, explain, settings
• Dynamic LLM Providers — Any OpenAI-compatible API, Claude, or local Ollama
• Dynamic Model Selection — Fetches available models from the API, no hardcoded lists
• RAM-Based Restrictions — Prevents selecting models too large for your hardware
• Auto-Install Ollama — Installs Ollama automatically if not present
• Smart Code Chunking — Splits large files at function boundaries with context memory between chunks
• Automatic Retry — Exponential backoff for transient failures (timeouts, rate limits, 5xx errors)
• Connection Pooling — Keep-alive HTTP connections for fast sequential LLM calls
• Command Generation — Natural language → shell commands, auto-copied to clipboard
• Safety Warnings — Flags dangerous commands before execution
• Cross-Platform — macOS, Linux (Ubuntu/Debian/etc.), Windows

curl -fsSL https://raw.githubusercontent.com/rgcsekaraa/niko-cli/main/install.sh | sh

iwr -useb https://raw.githubusercontent.com/rgcsekaraa/niko-cli/main/install.ps1 | iex

# Install latest version from git
cargo install --git https://github.com/rgcsekaraa/niko-cli

# Or install from local source
cargo install --path .

# First run — interactive setup wizard
niko settings configure

This will:

1. Show available providers (Ollama, OpenAI, Claude, DeepSeek, Grok, Groq, Mistral, Together, OpenRouter, or custom)
2. For Ollama: auto-install if needed → list local models → show downloadable models filtered by your RAM → let you pick
3. For API providers: ask for API key → fetch available models from the API → let you pick
4. Save everything to ~/.niko/config.yaml

$ niko cmd "find python files modified today"
find . -name "*.py" -mtime 0
Copied to clipboard

$ niko cmd "kill process on port 3000"
$ niko cmd "compress logs folder to tar.gz"
$ niko cmd "git commits from last week"
$ niko cmd "show disk usage by directory"

# From a file
niko explain -f src/main.rs

# Pipe code in
cat complex_module.py | niko explain

# Paste interactively (live line counter, Ctrl-D or two empty lines to finish)
niko explain

For large files, Niko:

1. Chunks code at function/block boundaries (max 200 lines/chunk)
2. Carries context — each chunk includes overlapping lines and a running summary from previous chunks
3. Retries failed LLM calls with exponential backoff (3 attempts, 500ms → 4s delay)
4. Synthesises chunk analyses into an overall summary with follow-up questions

# Interactive setup wizard
niko settings configure

# Show current config
niko settings show

# Set a value directly
niko settings set openai.api_key sk-xxx
niko settings set openai.model gpt-4o
niko settings set active_provider openai

# Reset to defaults
niko settings init

# Print config path
niko settings path

niko cmd "list files" --provider openai
niko explain -f main.rs --provider claude

Niko is designed for production use with reliability and speed:

All API providers fetch models dynamically from their /models endpoint — nothing is hardcoded.

API keys can also be set via environment variables:

export OPENAI_API_KEY=sk-xxx
export ANTHROPIC_API_KEY=sk-ant-xxx
export DEEPSEEK_API_KEY=xxx
export GROK_API_KEY=xxx
export GROQ_API_KEY=xxx
export TOGETHER_API_KEY=xxx
export MISTRAL_API_KEY=xxx
export OPENROUTER_API_KEY=xxx

For local models (Ollama), Niko estimates the maximum model size your system can handle:

Models exceeding your RAM limit are hidden from the selection list. You can still force-select them with a confirmation prompt.

All settings are stored in ~/.niko/config.yaml. The file uses a dynamic structure — providers are a map, so you can add as many as you want:

active_provider: openai
providers:
  ollama:
    kind: ollama
    base_url: http://127.0.0.1:11434
    model: qwen2.5-coder:7b
  openai:
    kind: openai_compat
    api_key: sk-xxx
    base_url: https://api.openai.com/v1
    model: gpt-4o
  claude:
    kind: anthropic
    api_key: sk-ant-xxx
    model: claude-sonnet-4-20250514

rm $(which niko)
rm -rf ~/.niko

MIT