AI Agent Hub
Back to plugins
🤖

dsh-reasoning-options

Model Inference Updated 2026.09.10

Run the following command in DeepSeek Harness:

dsh plugin install Scorp1o117/dsh-reasoning-options

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

Run dsh plugin install Scorp1o117/dsh-reasoning-options in the DeepSeek Harness terminal to install this plugin (source: https://github.com/Scorp1o117/dsh-reasoning-options).

About this plugin

Models configured through a pi-ai gateway miss two things that built-in DeepSeek models get for free: a reasoning-effort picker in the Web UI and the x-opencode-session header that OpenCode Go strictly requires for routing and prompt-cache management. Without the former you cannot toggle the seven intensity levels from off to max; without the latter every OpenCode Go request fails with a 400 MissingSessionID error.

dsh-reasoning-options closes both gaps at boot. It scans the llm-pi-ai namespace, writes the full seven-level reasoning declaration for any model that lacks one, defaults reasoning to high where no default is set, and injects the session header into opencode.ai provider routes. Every mutation flows through dsh's native schema-validated settings pipeline with hot-apply, so no file editing is involved. The plugin is idempotent: already-declared models and headers are left untouched, and providers added after install are picked up automatically via the settings/updated event.

It is built for anyone running gateway models through DeepSeek Harness that go via a third-party relay and want the same reasoning control as the native DeepSeek adapter. The plugin does not judge whether a model actually supports a given level; the choice is always yours.

Use Cases

  • Auto-completes the x-opencode-session routing header on OpenCode Go gateway models to prevent 400 MissingSessionID errors
  • Selects from seven reasoning-effort levels (off to max) in the Web UI for third-party pi-ai models without manual config edits
  • Newly added pi-ai providers are auto-discovered via the settings/updated event and receive reasoning declarations without a restart

Best For

  • DeepSeek Harness users running gateway models via OpenCode Go, GOAT, or Volcengine Ark
  • Developers who want third-party gateway models to match the reasoning control of native DeepSeek models
  • Ops engineers who prefer automated injection of gateway-required headers over manual settings.yaml edits