dsh-orchestrator

In AI agent workflows, multi-agent collaboration often faces challenges like complex communication, reliance on polling or conventions, leading to high development and maintenance costs. The dsh-orchestrator plugin addresses this by providing a native all-to-all communication solution for DeepSeek Harness. It enables a main agent to decompose and dispatch tasks, while multiple workers handle execution, with direct messaging between any agents—eliminating the need for file inboxes or prompt conventions and streamlining collaboration.

The plugin's core strength lies in its seamless integration and flexibility. It leverages DeepSeek Harness's built-in agent inbox system for direct message delivery, supporting any communication pattern like main-to-worker, worker-to-main, and worker-to-worker. Additionally, each worker can be individually configured with a specific model, max tokens, and thinking intensity, allowing for fine-grained control over task assignment and execution. This design makes it easy to build complex, dynamic agent meshes.

dsh-orchestrator is ideal for developers and researchers using DeepSeek Harness, especially those building multi-agent collaboration systems, automated task flows, or exploring agent interaction patterns. Whether you aim to simplify existing workflows or develop new agent applications, this plugin provides efficient and reliable orchestration support, letting you focus on core logic rather than communication details.