DeepSeek Harness setup and Claude Code comparison

Learn how the DeepSeek Harness uses API models or Ollama, adds memory and drafts Gmail messages through Zapier, plus its limits versus Claude Code.

Player not loading? Watch on YouTube

This walkthrough presents the DeepSeek Harness as a configurable coding assistant that runs on your machine. The speaker describes it as free, open source and MIT licensed, with model selection separate from the harness that handles tools and the interface. Setup involves copying the GitHub repository link and asking Claude to download it and open a localhost instance; the video does not give a detailed manual installation procedure.

The examples cover building a memory tool with save, search and forget functions, then connecting Zapier to draft a Gmail message about Budapest. The speaker uses Zapier as a shared authentication layer for services. He describes standard, PTC and creative modes, briefly naming minimal mode. PTC combines tool operations in a TypeScript program.

For people who want to run models locally, the speaker says the harness supports Ollama. API providers, OpenRouter, Zapier and Gmail involve hosted services; local harness execution is separate from local inference, and no Ollama setup is demonstrated. The harness itself has no charge, but paid model tokens still cost money. He says it requires API keys rather than Claude or Codex subscriptions and warns that the early software may break.

The comparison favors Claude Code or Codex for polished client work and DeepSeek for volume tasks where cost matters. These are the speaker's judgments, rather than benchmark results established by the walkthrough.