Run Claude Code Locally with Docker Model Runner

How to Build an App With Claude Code - Full Tutorial for Beginners

A how-to explains running Anthropic-compatible Claude Code locally by using Docker Model Runner to host open-source LLMs and point Claude’s CLI at the local API. The guide walks through prerequisites (Docker with Model Runner), pulling a model (example: ai/phi4:14B-Q4_K_M), checking model status with docker model commands, testing the /v1/messages endpoint via curl, and configuring Claude Code by setting ANTHROPIC_BASE_URL to the local TCP endpoint. The post highlights reasons to run locally—cost, privacy, and offline use—and shows persisting the environment variable in shell config. This lets developers use Claude’s tooling while relying on local models for coding workloads.

Developer guide shows how to run local LLMs with Docker Model Runner and point Claude Code at the local endpoint to avoid cloud tokens, costs, or data exposure. It walks through prerequisites (Docker Desktop/Engine, Model Runner, Claude Code), pulling a model from Docker Hub (example ai/phi4:14B-Q4_K_M), verifying model status with docker model ls/status, testing the HTTP API via curl against localhost:12434/v1/messages, and setting ANTHROPIC_BASE_URL to route Claude Code to the local model. The piece also recommends persisting the env var in shell config so the local endpoint remains available across sessions. This enables offline, private, and cheaper code-focused LLM usage.