deepseek-harness-model-config

DeepSeek Harness Model Configuration Plugin

English | 中文

Adds an Advanced Model Configuration page to the DeepSeek Harness Web UI. Use it to create custom model endpoints and configure each model's capabilities.

Features

• Create custom endpoints with a name, URL, API key, and protocol.
• Configure provider-level custom request headers for model requests.
• Supports openai-completions, openai-responses, and anthropic-messages.
• Fetch candidate models through a unified GET /models flow, search by model ID or display name, and select all, invert selection, or select none.
• Add a model manually by entering its model ID without first fetching; entering an existing ID selects that model instead of duplicating it.
• Use the same editor for new and saved endpoints; saved endpoints can refresh models, new models start unchecked, and checked models remain selected after a refresh.
• Endpoint advanced parameters cover default capacities, input modalities, reasoning compatibility, caching, transport, timeouts, and retry policies.
• Completes missing context windows, maximum output, input modalities, and reasoning capabilities from models.dev; it first selects the official provider implied by the display name (falling back to the model ID), then falls back to the default provider record, with whole-record switching available before saving.
• Preserves the exact display-name spelling returned by the endpoint or entered by the user, and rematches metadata providers after a display-name edit.
• Edit each selected model's capacity, text/image input support, and reasoningEfforts; review a configuration preview before saving.
• Chinese and English UI copy follows the Harness language setting.
• API keys are written only through Harness credential storage, never to settings.yaml or the configuration preview.

See the llm-pi-ai parameter reference for the complete configuration table.

Open settings

Add an endpoint Enter the endpoint name, BASE_URL, and API key, then fetch models and check the ones you want:

Fill in remaining parameters

1. Endpoint advanced parameters: the defaults are usually fine. Custom headers apply only to model requests; Fetch available models still uses the API key.

2. Model parameters: mainly check the context window and maximum output length.

Model parameters are usually filled in automatically, as long as https://models.dev is reachable.

Install

Install and run dsh, then install from GitHub by default:

dsh plugin --profile web add github:MarvekG/deepseek-harness-model-config
dsh web

Open Settings → Advanced model config in the Web UI to create endpoints or edit existing llm-pi-ai model configuration.

For a reproducible version, append a commit SHA to the repository address, such as github:MarvekG/deepseek-harness-model-config#<sha>.

Local Debugging

After cloning this repository, install the local link: dependency from its root:

dsh plugin --profile web add .
dsh web

Uninstall

Remove the plugin from the Web profile:

dsh plugin --profile web remove dsh-models-config-plugin

Update

Update by removing the old version and adding the new one:

dsh plugin --profile web remove dsh-models-config-plugin
dsh plugin --profile web add github:MarvekG/deepseek-harness-model-config
dsh web

For local debugging, replace the second command with dsh plugin --profile web add ..

License

MIT

Links

• Linux DO