DeepSeek Harness
DeepSeek Harness
DeepSeek Harness (the dsh command) is DeepSeek's open-source agent harness. Connect it to 上上之钥 through its Custom model API and a single click on Fetch available models lists every model your key can use, with the display name, context window and max output tokens already filled in.
1. Get an API key
Create a key on the API Keys page of the console. The key's group decides which models it can use:
2. Start DeepSeek Harness
Your browser opens http://127.0.0.1:3080 (pass --no-open to skip that).
3. Add a custom model API
Open Settings → Models, click Custom model API and fill in:
Pick the base URL and protocol by the key's group:
4. Fetch the model list
With the base URL, protocol and key filled in, click Fetch available models; In Choose models to add, tick the models you need (search is available) and click Add selected; Save the provider, then pick one of its models in the session's model selector.
The 国产模型 group also includes image, video and speech models. DeepSeek Harness is a chat agent, so pick text models only, for example deepseek-v4-pro, deepseek-v4.1-flash, kimi-k3 or glm-5.3.
Every fetched model comes with:
Name, e.g. DeepSeek V4 Pro; Context window, e.g. 1,000,000 for the DeepSeek V4 family; Max output tokens, e.g. 384,000 for the DeepSeek V4 family.
If a model arrives without a context window, edit it in the model row. Left empty, DeepSeek Harness assumes 262,144, which may be smaller than the model supports.
FAQ
The list shows Claude models but no DeepSeek? The key is not in the 国产模型 (Chinese models) group. Use a key from that group instead.
401 or check the API key? Check that the key is correct and not disabled or expired, and that the base URL matches the table above. With the Anthropic Messages protocol, leave /v1 off the URL.
Can I use the built-in "DeepSeek" provider instead? Not recommended. The built-in DeepSeek provider targets DeepSeek's official API, sends official-only extensions and uses a fixed model list. Use Custom model API to connect to 上上之钥.
Where do I see usage and cost? On the Usage page of the console, filtered by key. Billing is the same as for any other client.