Morgan Linton releases Small Harness v0.5 on GitHub, a local terminal-based agent harness optimized for running token-efficient overnight builds using small language models.
Morgan Linton announced the release of Small Harness v0.5 on GitHub. Small Harness is a terminal user interface (TUI) agent harness designed to run small language models (LLMs) locally on macOS. The tool supports multiple backends including Ollama, LM Studio, MLX, llama.cpp, and OpenRouter, offering a safe environment with approval gates for file and shell commands. The v0.5 release aims to enable developers to run overnight builds in a token-efficient, quality-optimized manner.
Running AI agents locally is becoming increasingly practical as smaller models grow more capable, and Small Harness addresses key developer pain points around cost, privacy, and execution reliability.
* Local execution eliminates API costs and data privacy concerns for repetitive tasks like overnight builds.
* Support for multiple backends gives developers the freedom to swap out underlying models seamlessly.
* Robust parsing and inline JSON detection are crucial for making smaller, less capable models (e.g., 7B parameters) work reliably with tool calling.
* Mandatory approval gates and diff previews provide necessary safety rails before executing commands locally.
DISCOVERED
5d ago
2026-06-07
PUBLISHED
5d ago
2026-06-07
RELEVANCE
AUTHOR
morganlinton