ochyai/vibe-local

What it solves

vibe-local provides a completely free, offline AI coding environment for Mac and Linux users. It removes the need for paid API subscriptions or an active internet connection, making AI-assisted programming accessible for students, beginners, and those in network-restricted environments like offline workshops.

How it works

The project acts as a launcher and configuration layer that integrates three main components: Ollama (for running local LLMs), a TUI (primarily using the OpenCode engine), and a set of optimized local models (such as Qwen and gpt-oss).

Key technical features include:

• Automatic Model Selection: It detects system RAM to automatically create Ollama model aliases (like vibe-coder or vibe-fast) with appropriate context window sizes (num_ctx) to prevent memory crashes.
• Intelligent Routing: A router (vibe-router) can split tasks between small models for quick chat and large models for complex coding.
• Flexible UI Options: Users can choose between the OpenCode TUI, a compatibility mode for the Claude Code CLI, or a built-in Python-based engine (vibe-coder.py).
• Local RAG: The built-in Python engine supports local Retrieval-Augmented Generation using sqlite3 and Ollama embeddings to inject codebase context into prompts.

Who it’s for

• Students and Educators: Specifically designed for non-profit research and education, particularly for those without paid AI plans.
• Beginners: People learning terminal operations through natural language.
• Privacy-Conscious Developers: Those who want a local-first, no-cost coding agent.

Highlights

• Zero Cost & Offline: Runs entirely on the user's hardware without network requirements.
• Classroom Mode: Allows a teacher to run a model server on one Mac and let students attach to it via LAN.
• RAM-Aware Configuration: Automatically optimizes model selection and context length based on available system memory.
• Multi-Engine Support: Supports multiple TUI interfaces and a standalone Python engine with XML tool-call fallback for models that lack native tool-calling capabilities.
• Hybrid Connectivity: An --auto mode that switches between cloud-based Claude Code (when online) and local LLMs (when offline).