Run local tool-calling agent inference with Rapid-MLX
Serve OpenAI- and Anthropic-compatible local LLM endpoints on Apple Silicon so coding agents can run tool-calling workflows against on-device models.
npx skills add agentskillexchange/skills --skill run-local-tool-calling-agent-inference-with-rapid-mlx
Use Rapid-MLX when an agent workflow needs a local Apple Silicon inference endpoint instead of a hosted model. The operator installs Rapid-MLX, serves a chosen MLX model at localhost:8000/v1, points Codex CLI, Claude Code, Aider, Cursor, Continue.dev, or another compatible agent runtime at that endpoint, and then runs coding, evaluation, or privacy-sensitive tool-calling sessions against an on-device model. Invoke this when code or data should stay local, when a team needs to test tool-calling behavior across local models, or when a Mac is being used as a repeatable inference appliance. The scope boundary is the agent-facing local inference endpoint and its runtime configuration, not generic model browsing, a Mac desktop app listing, or broad LLM infrastructure.
What this skill actually does
Inputs and prerequisites: Apple Silicon Mac, Rapid-MLX, MLX-compatible model, OpenAI- or Anthropic-compatible agent client such as Codex CLI, Claude Code, Aider, Cursor, or Continue.dev.
Setup notes: Install Rapid-MLX using the documented installer or package path, start the server with rapid-mlx serve, confirm the OpenAI-compatible endpoint at localhost:8000/v1, then configure the target agent client to use that local base URL and model alias.
Source and verification boundary: use https://rapidmlx.com/docs/ as the canonical reference before running the workflow; keep commands, API calls, CLI usage, and generated outputs reviewable against that upstream source.
Framework fit: publish this as a Multi-Framework workflow only when the operator can invoke the documented toolchain directly, rather than treating the upstream project as a generic product listing.