DeepSeek Harness Setup: A Free Claude Code You Own In 10 Minutes

DeepSeek Harness Setup: A Free Claude Code You Own In 10 Minutes

You can subscribe to my newsletter so you never miss a drop: https://ift.tt/fjuNc02 DeepSeek Harness Link: https://ift.tt/HYbmAj5 DeepSeek Harness is an open-source coding agent people are calling a free Claude Code alternative. In this walkthrough, I install it from scratch, connect a model, build a custom plugin, trace every tool call, and deliberately break a live run to see whether it can recover. The biggest idea is not DeepSeek versus Claude Code. The model is only the brain. The harness controls the context, tools, plugins, MCP servers, permissions, memory, and execution around it. DeepSeek made that layer free, open, and customizable, while still letting you connect DeepSeek, GLM, Ollama, local models, Anthropic, OpenAI, Claude Code, or Codex. You will also see the important limitations. The harness is still a developer preview, breaking changes are expected, hosted free models have limits, and truly uncapped free usage requires running an open model locally. Chapters: 0:00 DeepSeek Harness in 10 Minutes 0:29 What an Agent Harness Actually Is 1:55 Install DeepSeek Harness 3:18 Run It Free With Local Models 4:56 Why Everything Is a Plugin 6:28 Trace Every Tool Call 7:17 Breaking a Live Run on Purpose 8:46 Building a Custom Plugin 9:51 DeepSeek Harness vs Claude Code 10:08 Run Claude Code and Codex as Subagents 11:27 The Move Nobody Is Talking About 12:16 Final Verdict Which model would you connect to DeepSeek Harness first?