DeepSeek Harness

Create a Custom model API provider in the DeepSeek Harness developer preview.

Install and prepare

  1. Install Node.js and launch the npm package below. DeepSeek Harness is a developer preview with breaking changes; check its official repository when your UI differs. The command opens a local browser UI; that local address is not the API endpoint.
  2. Create a routerra.ai key on API Keys and copy a tool-capable model ID from Models.
npx @deepseek-ai/dsh web

Create the custom model API

  1. Open Settings → Models → Add model provider → Custom model API. Enter the fields below and paste your routerra.ai key into API key. Use Fetch available models → Add selected, or enter YOUR_MODEL_ID manually when discovery is unavailable, then save. A provider has one protocol: do not mix Chat Completions and Responses models under it.
  2. Select the saved model in the composer and create a new session. Existing sessions that already sent a request retain their previous model. Model settings apply to the next request without restarting the server.
Provider ID: routerra
Display name: routerra.ai
Base URL: https://routerra.ai/v1
API protocol: openai-completions
API key: YOUR_API_KEY
Model ID: YOUR_MODEL_ID

Check the connection

MISSING_CREDENTIAL: re-save the key in Models. UNKNOWN_MODEL: add the exact ID to this custom provider. Empty discovery does not prevent manual setup. Configuration is written to $DSH_HOME/profiles/web/cordis.patch.yml; the secret is stored separately in $DSH_HOME/.credentials.yaml. If the server rejects developer or max_completion_tokens, check the provider guide’s compat.supportsDeveloperRole and compat.maxTokensField options against the selected model before changing them.