DeepSeek Harness

1. Get a SupaNexus API key

Sign in to the Developer Console, open your project → API Keys, create a key, and copy the secret immediately (shown once). One project key works across OpenAI-compatible and Anthropic-compatible clients.

2. Install DeepSeek Harness

Install from the official developer preview and pin a version you have tested — the preview still ships breaking changes. The bundled DeepSeek catalog card talks to DeepSeek’s own API; you need a custom provider to use SupaNexus.

3. Add a custom provider

Open Settings → Models → Add a custom provider. Set a lowercase Provider ID such as supanexus (permanent for sessions and credentials). Base URL: SupaNexus host + /v1. API protocol: OpenAI Chat Completions (openai-completions). Paste your project API key.

Provider ID: supanexus
Base URL: <BASE_URL>/v1
API: openai-completions
API Key: sk-snx-...

4. Add models (or fetch the catalog)

Add IDs such as deepseek/deepseek-v4-pro and deepseek/deepseek-v4-flash, or use Fetch available models (GET /v1/models). Qwen, Kimi, GLM, and other Public API IDs work the same way — not only DeepSeek.

Recommended:
deepseek/deepseek-v4-pro    # flagship / agents
deepseek/deepseek-v4-flash  # latency / cost

5. Optional: edit settings.yaml

For fields the form does not expose, edit $DSH_HOME/settings.yaml and restart. Keep the secret in an env var (e.g. SUPANEXUS_API_KEY), not in YAML.

llm-pi-ai:
  providers:
    supanexus:
      apiKeyEnv: SUPANEXUS_API_KEY
      api: openai-completions
      baseURL: "<BASE_URL>/v1"
      models:
        - id: deepseek/deepseek-v4-pro
        - id: deepseek/deepseek-v4-flash

6. Pick a catalog model ID

Use exact IDs from the SupaNexus Models marketplace or GET /v1/models (for example deepseek/deepseek-chat, anthropic/claude-3-5-sonnet). Do not invent vendor-only slugs from other gateways.

7. Verify and smoke-test

Save, pick a SupaNexus model in Harness, and send a short prompt. Confirm usage in the Developer Console. Use the official DeepSeek catalog card only when you intend to call DeepSeek’s own API.

8. Troubleshooting

Still hitting DeepSeek cloud: you are on the bundled DeepSeek card — switch to the custom provider. 401 / wrong host: Base URL must include /v1. Model missing: ID must exist in GET /v1/models. Prefer Chat Completions for new Harness setups; only point at Messages if the rest of your stack already speaks Anthropic.