To everyone using still ollama/lm-studio... llama-swap is the real deal

A new open-source tool called llama-swap is gaining traction as a powerful alternative to popular local model servers like Ollama and LM Studio. Created by developer mostlygeek, llama-swap distinguishes itself by being provider-agnostic—it can work with any underlying inference engine, including llama.cpp, ik_llama.cpp, and planned image generation backends. Its core appeal lies in its extreme simplicity (a single binary and config file) paired with advanced features like live configuration watching, model grouping, policy enforcement, and a built-in web UI for testing and debugging. This combination addresses a key pain point for developers who need to manage and serve multiple AI models locally but find existing solutions either too rigid or resource-heavy.

The technical implementation is designed for seamless integration into developer workflows. It can be configured to start on boot via systemd user services, consuming minimal resources when idle. A standout feature is its powerful configuration file, which allows users to define macros, group models, force specific parameters (like temperature for certain tasks), and apply filtering rules. This makes it particularly useful for agentic workflows, where different models might be routed based on the task. The project is available on GitHub with clear setup instructions for Linux systems, suggesting a focus on server and development environments where efficient, multi-model orchestration is critical.