DeepSeek Harness
A plugin-based agent framework for developers, with a Web UI and a headless runner
About DeepSeek Harness
DeepSeek Harness is a framework for building agent runtimes rather than a ready-to-use coding-agent product. It provides a runnable Web UI and a headless runner, and organizes models, tools, skills, sessions, sandboxes, storage, loops, scheduling, and the interface as plugins that can be selected, replaced, or extended. The framework also provides models with a workspace, tool registry, sandbox, approval policies, append-only session logs, and loops for continuing work.
Best for
Best use cases
Select models, tools, and skills through configuration
Run agents through the Web UI
Execute tasks with the headless runner
Develop or extend agent plugins
Features
Plugin ecosystem
Models, tools, skills, sessions, sandboxes, storage, loops, scheduling, and the interface are all provided by plugins.
Runtime interfaces
Provides a runnable Web UI and a headless runner.
Agent core
Provides models with a workspace, tool registry, sandbox, approval policies, append-only session logs, and loops for continuing work.
Model integration
Supports connecting to DeepSeek through a single API key field, using catalog providers, or configuring a custom OpenAI-compatible endpoint.
Pros
- Plugins cover multiple capability categories, including models, tools, skills, and sessions
- Provides both a Web UI and a headless runner
- Supports selecting, replacing, or extending plugins through configuration
- Released under the MIT License with the full source code publicly available
Cons
- Still in developer preview
- The repository warns that breaking compatibility changes may occur