DeepSeek releases plugin-based Harness in developer preview
DeepSeek has released DeepSeek Harness v0.1 as a developer preview, with its source code available under the MIT license. Distributed as dsh in the deepseek-ai/deepseek-harness project, it is designed as infrastructure for assembling agent runtimes rather than as a finished coding assistant. The system places models, tools, skills, sessions, sandboxes, storage, control loops, scheduling and the user interface behind Cordis plugin boundaries.
Harness offers four runtime modes. Standard provides a full coding-agent environment with file and web search, shell access, planning, goals, subagents and workflows. Code mode exposes tools through the Code Mode SDK, Minimal keeps persistent bash and str_replace_editor for model benchmarking, and Creator supports runtime inspection and plugin experiments.
Each run is recorded in an append-only session log containing the model’s context, tool activity, results, subagent scheduling and context injections. Resume, fork, search and replay use the same event stream. Model routing is provider-agnostic, supporting DeepSeek, Anthropic, OpenAI, Bedrock, Vertex, Azure, Codex and OpenAI-compatible endpoints. The web interface starts with npx @deepseek-ai/dsh web and normally runs at http://127.0.0.1:3080. A Python SDK requires Python 3.10 or newer.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.