DeepSeek Harness is an open-source AI coding agent environment built by DeepSeek for developers who want a configurable alternative to OpenAI Codex and Anthropic Claude Code. Released on August 13, 2026 under the MIT license, it turns the model's reasoning into real engineering actions: reading files, calling tools, writing code, and running tests. You install it with npm install -g ds-h and launch the web profile to begin.
Core Features
- Plugin-everything architecture: model connections, tool calls, session logs, and human approvals are all swappable plugins, so any module can be replaced or extended.
- Four preset agent modes: Standard, PTC for TypeScript automation, Minimal, and Creative, covering coding assistants through content generation.
- Multi-model support across roughly 40 providers including OpenAI, Anthropic, Google, and Kimi, so you are never locked to one backend.
- Native terminal workflow with approval gates, plus about 300 community plugins already built during the private beta.
Use Cases / Best For
- Engineers who want a local-first, auditable coding agent instead of a closed SaaS product.
- Teams standardizing on DeepSeek V4 Pro for agentic coding with OpenAI Responses API compatibility.
- Platform builders packaging coding agents for their own users via the plugin SDK.
Pricing
DeepSeek Harness is free and open-source under the MIT license. There is no subscription; you pay only for the underlying model API tokens you choose to connect.
Pros & Cons
- Pros: fully open-source and self-hostable; the plugin architecture is unusually flexible.
- Cons: you must supply and manage your own model API keys and plugins.
Our Take: Best for developers who want a transparent, self-hostable Claude Code alternative and do not mind wiring up their own model keys. The trade-off is that, unlike hosted products, setup and plugin curation are on you. See also our coverage of Grok Build and the AI programming category.
