DeepSeek Harness

Connect DeepSeek Harness to AIHubMix: install the npm release, configure an OpenAI-compatible provider, add auto or specific models, and test an Agent request.

DeepSeek Harness (dsh) is DeepSeek’s open-source coding agent harness. It provides workspaces, a terminal, file editing, Skills, planning, and a Web UI. Its OpenAI Chat Completions compatibility lets it connect to AIHubMix.

This guide is based on DeepSeek Harness 0.2.0-rc.2, the latest version as of October 18, 2026. This version differs significantly from 0.1.0-rc.6, and the configuration location and UI have been updated. Harness is still a developer preview, so later releases may change fields or UI details.

​

Quick summary

The integration requires four values:

Setting	Value
Provider ID	aihubmix
Base URL	https://aihubmix.com/v1
API protocol	openai-completions
API key environment variable	AIHUBMIX_API_KEY

After configuration, the Models page shows AIHubMix as configured.

​

What do you need?

Before you begin, prepare:

• An active AIHubMix API key
• Node.js 22 (22.19.0 or later) or Node.js 24 or later
• Network access to the npm registry and https://aihubmix.com

Check Node.js and npm:

node --version
npm --version

​

How do you install DeepSeek Harness?

Install the published npm package without cloning the source repository:

npm install -g @deepseek-ai/dsh

Verify the installation:

dsh --version

If the shell reports dsh: command not found, run npm config get prefix and make sure its bin directory is included in PATH.

​

How do you start the Web UI?

Run the following command from the directory you want to use as the workspace. The officially recommended way to start Harness is npx:

npx @deepseek-ai/dsh web --port 3080

If Harness is already installed globally, you can also run:

dsh web --port 3080

Open this URL:

http://127.0.0.1:3080

On first use, select or add a workspace before sending an Agent request.

​

How do you configure the AIHubMix provider?

Open Settings → Models → Add model provider, switch the card to Custom model API, then enter the following values:

Field	Recommended value	Notes
Provider ID	aihubmix	Lowercase identifier; use aihubmix-gateway if it is already taken
Display name	AIHubMix	Name shown in the model selector
Base URL	https://aihubmix.com/v1	Must include /v1
API protocol	openai-completions	OpenAI Chat Completions compatible protocol
API key	Your AIHubMix API key	Harness stores it as a credential and does not show the plaintext value again

Harness appends /chat/completions to the Base URL. The resulting endpoint is:

https://aihubmix.com/v1/chat/completions

Do not put the full /chat/completions path in the Base URL field.

​

Add auto or another model

After entering the Base URL and API key, open Model catalog and select Fetch available models. Harness requests GET https://aihubmix.com/v1/models and shows the currently available model IDs.

For an initial connection test, you can add only auto and set its display name to Auto. auto is AIHubMix’s automatic model selection entry point.

To use a specific model, select it from the fetched list, then select Add selected. The graphical model selector only shows models registered in the current provider catalog.

​

auto vs specific models

Choice	When to use it
auto	Initial connection tests, or automatic model selection by AIHubMix
Specific model ID	A fixed model capability, version, or reproducible result is required

Both choices use the same provider, Base URL, and API key. Switching models does not require creating the provider again.

​

Store the API key in an environment variable

For development machines and CI, Harness can read the key from an environment variable through a profile file. The Web UI uses the web profile, and headless tests use the headless profile. The file paths are:

• $DSH_HOME/profiles/web/cordis.patch.yml
• $DSH_HOME/profiles/headless/cordis.patch.yml

Both files use the same content:

- id: llm-pi-ai
  config:
    providers:
      aihubmix:
        apiKeyEnv: AIHUBMIX_API_KEY
        api: openai-completions
        baseURL: https://aihubmix.com/v1
        models:
          - id: auto
- id: agent-default-model
  config:
    provider: aihubmix
    model: auto

Both entries are required. If you write only llm-pi-ai and omit agent-default-model, headless still uses the default deepseek-official provider and reports MISSING_CREDENTIAL.

Earlier versions stored the configuration in $HOME/.dsh/settings.yaml, which version 0.2.0 no longer uses. When DSH 0.2.0 starts for the first time, it migrates that file to a profile configuration and renames the original file to settings.yaml.imported. The migration run may report MISSING_CREDENTIAL. Run the command once more.

Set the key before starting Harness:

export AIHUBMIX_API_KEY='sk-***'
dsh web --port 3080

When the key comes from the launch environment, the settings page displays a read-only message without revealing the value.

Restart Harness after changing the environment variable.

​

Select a model and test a conversation

Close Settings, open the model selector at the lower right of the prompt box, and choose Auto or another model you added.

Send a minimal test message:

Reply with exactly: AIHubMix connection successful

Receiving the expected response confirms the connection between Harness, the OpenAI Chat Completions compatible endpoint, and AIHubMix.

​

Run a headless test

You can also test without the Web UI. This command uses the headless profile, so $DSH_HOME/profiles/headless/cordis.patch.yml must contain both entries above.

export AIHUBMIX_API_KEY='sk-***'
dsh --profile headless \
  'Reply with exactly DSH_AIHUBMIX_OK and do not use tools.'

A successful run prints DSH_AIHUBMIX_OK.

​

FAQ

​

Why does Harness show MISSING_CREDENTIAL?

Harness did not read a usable configuration. Confirm that both entries are written to the profile, confirm that the AIHUBMIX_API_KEY environment variable is set, and then launch Harness from the same terminal:

test -n "$AIHUBMIX_API_KEY" && echo configured

​

Why does model discovery return 401?

Check that the API key is valid and does not contain extra spaces or line breaks. The model discovery request uses Bearer authentication.

​

Why is Auto the only model in the selector?

The model selector shows only models registered in the current provider catalog. Open Settings → Models → AIHubMix → Edit, select Fetch available models in Model catalog, choose the models you need, and then select Add selected.

​

Why does Harness report UNKNOWN_MODEL?

The current provider catalog does not contain that model ID. Model IDs are case-sensitive, so use the exact value returned by model discovery.

​

Why does the connection time out?

Check DNS, your organization network, and local proxy settings. DSH reads the HTTP_PROXY and HTTPS_PROXY environment variables automatically. To exclude specific addresses from the proxy, set NO_PROXY. If the browser uses a system proxy but the terminal does not, configure HTTP_PROXY and HTTPS_PROXY in the terminal that launches dsh.

​

Security guidance

• Do not commit API keys to Git or include them in screenshots and shared scripts
• Prefer Harness credential storage or the AIHUBMIX_API_KEY environment variable
• Do not print a complete API key while troubleshooting
• Use the CI platform’s secret management feature to inject environment variables

---

Last updated: 2026-10-18