sahellx/kitty-bash-llm

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 31, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

sahellx/kitty-bash-llm is a 0.5 billion parameter shell assistant model, built on Qwen2.5-Coder-0.5B-Instruct, designed to run efficiently on laptop CPUs. It specializes in converting natural language to bash commands, fixing broken commands, and completing partial commands. Its key differentiator is its small size (398 MB GGUF) and rapid inference speed (under 1 second on CPU), enabling local, offline use without a GPU or API key. This model is a specialist, excelling in shell command generation and repair, rather than general chatbot capabilities.

Loading preview...

Kitty Bash LLM: A Compact Shell Assistant for CPU Inference

Kitty Bash LLM, developed by sahellx, is a 0.5 billion parameter model specifically engineered to assist with bash commands. Built on the Qwen2.5-Coder-0.5B-Instruct base, this model is designed for extreme efficiency, with its 4-bit GGUF version weighing only 398 MB, allowing it to run on a laptop CPU in under a second per command. This eliminates the need for a GPU, API keys, or network round trips, making it ideal for seamless shell integration.

Key Capabilities

  • Natural Language to Command (nl2cmd): Translates plain English requests into precise bash commands (e.g., "find all files larger than 100MB" -> find /var -type f -size +100M).
  • Command Repair (fixcmd): Corrects broken bash commands based on error output (e.g., gerp -r 'TODO' . -> grep -r 'TODO' .).
  • Command Completion (complete): Autocompletes partially typed bash commands (e.g., tar -czf backup. -> $(date +%F).tar.gz /path/to/dir).
  • Specialized Focus: Deliberately designed as a specialist for shell tasks, not a general-purpose chatbot, ensuring high accuracy and relevance for command-line operations.
  • Optimized for CPU: Achieves fast inference times (around 700 ms for warm requests on a 2-thread CPU) by leveraging llama.cpp and GGUF quantization, making it suitable for persistent server setups.

Good For

  • Developers and system administrators needing an offline, local, and fast bash command assistant.
  • Integrating into shell environments for real-time command generation, correction, and completion.
  • Use cases where privacy and data locality are paramount, as it runs entirely on local hardware.
  • Users who prioritize efficiency and minimal resource consumption over general-purpose LLM capabilities.

⚠️ Safety Note: This model will generate destructive commands without hesitation. Any tool built on it must implement safety measures like never auto-executing, pattern-matching destructive commands, and treating output as untrusted suggestions.

Finetunes
1 models