---
title: "Connect DeepSeek Harness to XiuRouter"
description: "Add or update a DeepSeek Harness provider and fetch models accessible to your XiuRouter key."
image: "https://docs.xiu.ai/og.png"
---

> XiuAI documentation index
> Full documentation index: https://docs.xiu.ai/en/llms.txt
> Check the index for current pages before reading.

# Connect DeepSeek Harness to XiuRouter

Add a provider named `xiurouter`, or edit the existing provider if that ID already exists. DSH obtains models from `/v1/models`; you do not need to preselect one on the XiuRouter website.

## Choose a setup method

- First setup: add `xiurouter` in Settings → Models.
- Existing setup: edit the existing connection, or retain your provider and add XiuRouter separately.
- Chat-assisted setup: unavailable; DSH settings store the provider and credentials.

## Before you start

The DSH web UI must open, you need a dedicated XiuRouter key, and the workspace must have only the permissions needed for the task.

## Configure the provider

1. Open **Settings → Models**.
2. Open `xiurouter` if it exists; otherwise select **Add a custom provider**.
3. Use `xiurouter` as Provider ID. Sessions and credentials refer to this ID; renaming requires a new provider.
4. Set Base URL to `https://router-api.xiu.ai/v1`.
5. Select `OpenAI Chat Completions` as API protocol.
6. Save the XiuRouter key.
7. Use **Fetch available models** to list models accessible to the key, then select at least one.
8. Save, choose the XiuRouter model in the model selector, and start a new session.

The saved key is write-only and is not returned as plaintext by the settings page. Do not put it in workspace files or plugin configuration.

## Verify

1. Send a short request and confirm the session uses `xiurouter` and the expected model.
2. Ask DSH to read and summarize a nonsensitive file without editing it.
3. Confirm the model, endpoint, and successful request in XiuRouter's request inspection view.
4. Only then expand to edits, image input, or broader workspace permissions.

Manually added models default to text-only input. Declare `input: [text, image]` only after verifying image input with both the model and current service source.

## Troubleshoot

| Symptom | Next step |
| --- | --- |
| Fetch available models returns `401` | Check the saved key; you can also manually add an exact model ID |
| `UNKNOWN_MODEL` | Add that same ID to the `xiurouter` provider and start a new session |
| Text works but tools fail | Use a model with verified tool support and keep workspace permissions unchanged while troubleshooting |

## Roll back

Restore the original provider and model in **Settings → Models**. Verify a new session uses them before deleting `xiurouter` and revoking its dedicated key.

## Sources and review date

Reviewed on 2026-08-31. This translation preserves that date. Basic setup requires a full reply and matching usage record; verify tools with the target model in the actual workspace.

Source: https://docs.xiu.ai/en/router/integrations/dsh/index.mdx