bamboostrip/dsh-llm-capabilities
Detects which thinking levels and context limits a model actually accepts, and configures them for self-hosted gateways.
Patches the fields the official llm-pi-ai settings page leaves unconfigurable for a custom provider, where contextWindow is editable and thinking levels are not — which is why the intensity slider can say the model provides no selectable levels. Detection fetches GET {baseURL}/models server-side with the credential resolved host-side so no key leaks to the browser, and parses the several shapes gateways use for reasoning support and context limits, then merges pi-ai's own catalog as a complementary source with the endpoint winning on yes or no. It also sets the x-opencode-session header for OpenCode Go, keyed by a hash of the conversation id so the same chat keeps the same session across restarts.
Install
dsh plugin --profile web add dsh-llm-capabilitiesThe npm line, which the README marks as published. It succeeds dsh-reasoning-efforts rather than replacing it abruptly: keep the old package if you already run it, since this one serves both the current and the legacy endpoints, but new installs should take this one. The settings page appears as Model Capabilities, and saving writes into llm-pi-ai under providers.<route>.models with a revision guard.
Compatibility
Requires dsh 0.1.1-rc.2 or newer and Node 22.13 or newer. Since 0.2.0 it leaves input strictly alone — the official UI declares input modalities itself now, and existing values are preserved untouched on every save. Re-detection never erases saved reasoningEfforts, so a configured model keeps its own levels, custom wire spellings included, whether or not the endpoint lists any signals. Models the endpoint reports and you have not configured appear with a not-configured badge and are added on save.
Details
- Repo: bamboostrip/dsh-llm-capabilities
- Category: Models & Providers
- Stars: 2
- Version: 0.1.5 (npm)
Alternatives
tianxia--/dsh-llm-local-token · yuioi666/dsh-plugin-model-capability · drscrewdriver/dsh-llm-openai-completions