AI Agent Hub
Back to plugins
dsh-sonar preview

dsh-sonar

Client Updated 2026.08.15

Run the following command in DeepSeek Harness:

dsh plugin install Howe829/dsh-sonar

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

Run `dsh plugin install Howe829/dsh-sonar` in your terminal to install the plugin; the full source code is available at https://github.com/Howe829/dsh-sonar

About this plugin

The DeepSeek Harness Web UI offers a powerful chat interface, yet visibility into context-window pressure, cache hit rate, token spend, and account balance is scattered across logs and console output. dsh-sonar fills that gap with a compact, draggable Fish launcher that, when clicked, reveals a full session dashboard without narrowing the conversation column.

The dashboard surfaces a color-coded context-window gauge, a donut chart breaking down uncached input, cache read/write, output, and reasoning tokens, a heuristic composition view of system prompts and tool schemas, per-turn model and tool timing, cache hit rate, and cumulative token counts. A precise total_balance readout refreshes after every completed turn. While the model is generating, lightweight bubbles rise around the launcher, giving the wait a small visual rhythm.

It is aimed at developers and teams who run long multi-turn conversations or tool-heavy workflows in the Harness Web UI and need a glanceable cost-and-capacity picture. As a strictly read-only plugin, it injects no tools, alters no prompts, writes no durable session events, and touches neither the request envelope nor KV-cache behavior, so it can be added or removed without any side effect on model output.

Screenshots

Use Cases

  • Monitor context-window pressure and token breakdown during long conversations
  • Evaluate cache hit rate to refine prompt and tool-chain design
  • Quickly verify account balance and cumulative usage after multi-turn tool orchestration

Best For

  • Developers running long conversations or tool-heavy workflows in the Harness Web UI
  • AI teams that track token cost and cache efficiency
  • Users who want usage visibility without altering model behavior