Rapid-MLX: model tool for Hermes Agent

Q: Does Rapid-MLX work with Hermes Agent?

Yes. The README says rapid-mlx launch can run Hermes, along with Codex, Claude Code, OpenCode, Pi, Crush or Goose, using a local model.

OpenAI- and Anthropic-compatible local LLM server for Apple Silicon, with a one-command launcher for Hermes

What Rapid-MLX does

Rapid-MLX serves local models on a Mac through OpenAI-compatible and Anthropic-compatible APIs. The project focuses on tool calling for coding agents and claims decode speeds up to 4 times faster than Apple's mlx-lm on the same weights, with a published comparison against oMLX and LM Studio on a Mac Studio. Those numbers are the project's own measurements.

The top of the README shows a launcher: rapid-mlx launch <agent> runs Codex, Claude Code, OpenCode, Pi, Hermes, Crush or Goose with a local model. Running the same command again later reuses the setup, and adding --remove deletes Rapid's private profile without changing the agent's existing settings. The first reply can be slow while a model downloads and loads. It installs on a stock Apple Silicon Mac with a shell script, and the README links per-agent guides on rapidmlx.com. Topics include continuous batching, speculative decoding and MCP.

Key features

OpenAI- and Anthropic-compatible local server built on MLX
rapid-mlx launch <agent> for Codex, Claude Code, OpenCode, Pi, Hermes, Crush and Goose
--remove flag that removes Rapid's private profile without touching existing agent settings
Focus on reliable tool calling for coding agents
Benchmark tables against oMLX and LM Studio in the README

When to use it

Pointing Hermes at a local Qwen model on an Apple Silicon Mac
Running coding agents offline without a hosted API
Testing tool-calling reliability of local models across several agents

Who it is for: Apple Silicon Mac owners who want to run Hermes and other coding agents on local models.

How it fits with Hermes Agent

Rapid-MLX is a general local model server. Hermes is one of the agents its launcher supports.

How to install Rapid-MLX

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

curl -fsSL https://rapidmlx.com/install.sh | bash
export PATH="$HOME/.local/bin:$PATH"

Requirements: An Apple Silicon Mac; the chosen coding agent must be installed separately

Note: The license field is not identified by GitHub, although the README describes the project as Apache 2.0.

FAQ

What is Rapid-MLX?

Rapid-MLX is an open-source LLM inference server for Apple Silicon, built on MLX, that exposes OpenAI-compatible and Anthropic-compatible APIs and targets tool calling for coding agents.

Does Rapid-MLX work with Hermes Agent?

Yes. The README says rapid-mlx launch <agent> can run Hermes, along with Codex, Claude Code, OpenCode, Pi, Crush or Goose, using a local model.

How do I install Rapid-MLX?

On a stock Apple Silicon Mac, run the install script from rapidmlx.com, restart the terminal or add ~/.local/bin to PATH, then install your agent and run rapid-mlx launch.