Mark RadarMARK RADAR
About
EN
Sign in

DeepSeek Open-Sources Harness Agent Framework

1 reports · First detected 2026-08-14 · Last active 2026-08-14

Competition in AI agents is shifting beyond model performance to the harness layer, which manages tool calls, files, sessions and task orchestration. DeepSeek Harness targets that layer with an open, model-agnostic foundation designed to compete with Anthropic’s Claude Code and OpenAI Codex. Its significance lies in giving developers more control over agent workflows and reducing dependence on a single AI provider, while extending DeepSeek’s challenge from model pricing and performance into developer infrastructure.

DeepSeek released DeepSeek Harness v0.1 as a developer preview on Aug. 13, 2026, and published the code under the MIT license. Powered by the Cordis meta-framework, it treats models, tools, skills, sessions, sandboxes, filesystems, loops, orchestration and the user interface as interchangeable plugins. The framework can connect to models from multiple providers rather than locking users into DeepSeek. A related report said the project approached 60,000 GitHub stars overnight, signaling unusually strong developer interest.

All Coverage

1 original reports

The Backstory

The history behind this event
DeepSeek Said to Test Harness AI Agent Against Claude Code2026-07-28 · 1 reports · similarity 0.84

DeepSeek is extending its ambitions beyond foundation models into agentic coding tools with Harness, a desktop AI agent designed to compete with Anthropic’s Claude Code. Such products can plan, write and revise code while interacting with development environments, making them a key battleground for model providers seeking deeper integration into software workflows. The project also signals intensifying competition between Chinese AI developers and established U.S. coding assistants.

DeepSeek plans to begin a limited test of Harness during the week of July 28, 2026, according to the latest report. Selected participants will be required to sign strict non-disclosure agreements. Development is being led by Cui Tianyi, described as a former Jane Street expert. DeepSeek has not disclosed the number of testers, a public launch date, pricing or the amount invested in the project.

Mark Radar|MARK RADAR
All times are in Taipei time (GMT+8)