AI Agent Hub
Back to plugins
🤖

dsh-swarm-router

Model Inference Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install r600a-code/dsh-swarm-router

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install r600a-code/dsh-swarm-router in the DeepSeek Harness terminal to add the multi-model swarm router plugin, source at https://github.com/r600a-code/dsh-swarm-router

About this plugin

When a batch contains quick answers, deep reasoning, coding, and long context tasks, sending everything to the same model wastes budget and quality. dsh-swarm-router treats tasks as rows and candidate models as columns, uses capability filters, weighted scores, and difficulty penalties to pick the best cell for each row, then dispatches parallel subagents pinned to that model.

Routing does not consume model calls, so savings go to parallel execution. The plugin also records real task feedback and usage stats, boosting or demoting models by success rate and quality so rankings adapt to actual performance instead of relying only on static catalogs.

It suits DeepSeek Harness users who need to route heterogeneous task batches across multiple models while tracking cost, quality matching, routing preview, feedback rankings, and token usage.

Use Cases

  • Route a batch of heterogeneous tasks to different models by difficulty
  • Automatically pick the right model tier balancing cost and quality
  • Track per-model success rate and token spend to adjust routing over time

Best For

  • Developers batch-dispatching diverse LLM tasks
  • Users balancing model cost against output quality
  • Teams building multi-model pipelines on DeepSeek Harness