Houtini LM

•
122
•

Offload routine coding and analysis tasks from Claude Code to local or cloud LLM servers, significantly reducing API token costs.

Houtini LM empowers developers to significantly cut down their Claude Code API token expenditures by intelligently delegating routine, bounded tasks to more cost-effective local or cloud Large Language Models. Built out of a need to reduce bills from extensive refactoring, this tool allows Claude to focus on complex architectural decisions and multi-file changes, while offloading boilerplate generation, code reviews, commit messages, and format conversions to a cheaper, private, and rate-limit-free LLM server. It acts as a smart intermediary, optimizing your AI assistant's workload and budget by ensuring the right model handles the right task.

Key Features

01Automatic model discovery and capability profiling from HuggingFace for intelligent task assignment.
02Supports structured JSON output with grammar-based sampling for guaranteed valid responses.
03Real-time performance tracking with TTFT, tokens per second, and cumulative session stats.
04Intelligent model routing based on task type, model capabilities, and per-family prompt hints.
05Delegates routine coding tasks (e.g., test stubs, code review, commit messages) from Claude to local or cloud LLMs.
0613 GitHub stars

Use Cases

01Leveraging local LLMs (LM Studio, Ollama) for private, free, and rate-limit-free AI assistance.
02Optimizing AI assistant workflows by offloading specific tasks to specialized local or cloud models.
03Reducing Claude Code API token costs for repetitive or bounded coding tasks.