Connect DeepSeek Harness to NixAPI: Custom Model API Setup
Install DeepSeek Harness, add NixAPI as a custom OpenAI-compatible provider, and verify chat and workspace tool calls.
DeepSeek Harness (DSH) is an open-source agent tool with desktop and web interfaces. This guide uses its custom model API support to configure NixAPI. Instructions were checked against official documentation on October 30, 2026; successful requests depend on your key permissions and the selected model’s capabilities.
1. Prepare your connection details
Create a key in the NixAPI console and copy a chat model ID with tool support from models and pricing.
| Field | Value |
|---|---|
| Provider ID | nixapi |
| Display name | NixAPI |
| Base URL | https://api.nixapi.com/v1 |
| API protocol | OpenAI Chat Completions |
| API key | Your NixAPI key |
| Model ID | An exact ID from the current model list |
A model ID differs from a display name. NIXAPI_MODEL_ID is a placeholder, and the pricing website is not an API endpoint.
2. Start DSH
Download the desktop application from the official product page. For the web interface, install Node.js and run the command from the official README:
npx @deepseek-ai/dsh web
The default local address is http://127.0.0.1:3080. Follow the URL printed by your terminal if it differs.
3. Add your provider
Open Settings → Models → Add model provider → Custom model API. Enter the values above, then add a model and save. Use the custom provider form rather than entering a NixAPI key into the default DeepSeek account card.
Select OpenAI Chat Completions for this recipe. Try Fetch available models, or enter the exact model ID manually if discovery fails. Discovery alone does not verify tool calls. See the official provider guide.
4. Test a workspace
Create an empty test directory, add and select it as a workspace, then start a new session using your NixAPI model. Send these prompts separately:
Reply only: NixAPI connection works.
List the files in the current workspace and summarize its structure. Do not modify files.
Check the execution trace for an actual tool call and returned result. The Web UI guide explains why a workspace must be selected before starting a task.
5. Troubleshoot
- 401: check the key, balance, model permissions, and saved provider settings.
- 404: use
https://api.nixapi.com/v1; do not append/chat/completionsyourself. - Unknown model: copy the exact ID from current pricing.
- Chat works but tools fail: confirm tool calling and streaming support for that model and protocol before using it for agent tasks.
For advanced settings, the current web profile uses $DSH_HOME/profiles/web/cordis.patch.yml, normally under ~/.dsh. Open the actual configuration from Settings and preserve existing providers. Older settings.yaml examples may not apply. The adapter reference documents credentials and protocols.