Claude Cowork 3P

Configure AIOHub as the third-party inference gateway for Claude Cowork.

Claude Cowork 3P is the third-party inference deployment mode for Claude Desktop. In this mode, the Cowork and Code tabs send model inference requests to a configured provider instead of Anthropic’s first-party inference service.

AIOHub supports the Claude Messages protocol. Configure Cowork 3P with AIOHub as a gateway inference provider.

Claude Cowork 3P is in public beta. Entry points and managed-configuration behavior may change with Claude Desktop releases. This page follows Anthropic’s current public 3P configuration keys.

Even when you use the AIOHub API, Claude Cowork 3P is only usable in countries or regions where Anthropic provides service. This excludes mainland China, Hong Kong, and Macao. You are responsible for resolving Claude Desktop, Claude account, and network access requirements.

Model scope

When you use the AIOHub API with Claude Cowork 3P, model access is not limited to Anthropic or Claude models. If the model is visible to the group assigned to your API key and AIOHub currently supports it through the Claude Messages-compatible gateway, you can also choose OpenAI, open-source, and other models.

If your API key includes the auto group, the Cowork 3P model selector can show multiple model families in the user interface. The actual list still depends on the models visible to the current API key; check the console or /v1/models first when you need an exact model ID.

Prerequisites

• Claude Desktop with Cowork 3P support installed
• An AIOHub API key
• A group assigned to the API key that can access Claude-compatible models
• A positive account balance

Minimum settings

To configure AIOHub, you only need two fields:

Field	Value
Gateway base URL	https://api.aiohub.org
Gateway API key	sk-your-api-key

Set inferenceGatewayBaseUrl to the gateway root. Do not add /v1. Cowork 3P calls POST /v1/messages itself; if the base URL is https://api.aiohub.org/v1, the client may call the wrong path.

Keep Gateway auth scheme and Model list at their defaults. If your API key includes the auto group, the model selector can show the models visible to the current API key.

Configure model web access

If you want models in Claude Cowork 3P to access the internet, configure Allowed egress hosts. The managed configuration key is coworkEgressAllowedHosts; it controls which hosts the Cowork sandbox can reach for web fetch, shell commands, and package installs.

Common configurations:

Goal	coworkEgressAllowedHosts
Allow all URLs	["*"]
Allow only selected hosts	["docs.aiohub.org", "api.github.com"]
Allow a domain and its subdomains	["*.example.com"]

Setting this to ["*"] disables the Cowork sandbox egress host filter, so model tools can reach any URL. This is convenient for personal testing; for team or production deployments, allow only the hosts you actually need.

coworkEgressAllowedHosts controls external website access from tools inside the Cowork sandbox. It is not the AIOHub inference gateway Base URL. The configured inference endpoint is allowed implicitly.

Use the in-app configuration flow

1. Open third-party inference configuration

   Open Claude Desktop, then go to Developer -> Configure third-party inference.

2. Choose Gateway

   Select Gateway as the inference provider. Use this mode for gateways that implement the Anthropic Messages API.

3. Enter AIOHub settings

   Fill in only these two fields:

   Field	Value
   Gateway base URL	https://api.aiohub.org
   Gateway API key	sk-your-api-key

   Keep Gateway auth scheme and Model list at their defaults. If your API key uses the auto group, the model selector can show multiple models visible to the current API key.

   If you want the model to access external websites, configure Allowed egress hosts. To allow any URL, set it to *. To allow only specific websites, list hostnames such as docs.aiohub.org or *.example.com.

4. Export configuration

   Export a macOS .mobileconfig profile or Windows .reg policy from the wizard. For single-machine testing, you can also use the local user configuration file.

5. Restart Claude Desktop

   Quit and reopen Claude Desktop. Seeing the Cowork and Code tabs and a Gateway sign-in option means 3P mode was detected.

Local user configuration example

If you are not deploying configuration through MDM, add enterpriseConfig to the local user config. This is useful for personal testing, not for distributing real API keys across a team.

macOS

Config file location:

~/Library/Application Support/Claude-3p/claude_desktop_config.json

Minimum example:

{
  "enterpriseConfig": {
    "inferenceProvider": "gateway",
    "inferenceGatewayBaseUrl": "https://api.aiohub.org",
    "inferenceGatewayApiKey": "sk-your-api-key"
  }
}

To allow model web access to all URLs, add:

{
  "enterpriseConfig": {
    "inferenceProvider": "gateway",
    "inferenceGatewayBaseUrl": "https://api.aiohub.org",
    "inferenceGatewayApiKey": "sk-your-api-key",
    "coworkEgressAllowedHosts": ["*"]
  }
}

Windows

Config file location:

%APPDATA%\Claude-3p\claude_desktop_config.json

Minimum example:

{
  "enterpriseConfig": {
    "inferenceProvider": "gateway",
    "inferenceGatewayBaseUrl": "https://api.aiohub.org",
    "inferenceGatewayApiKey": "sk-your-api-key"
  }
}

To allow model web access to all URLs, add:

{
  "enterpriseConfig": {
    "inferenceProvider": "gateway",
    "inferenceGatewayBaseUrl": "https://api.aiohub.org",
    "inferenceGatewayApiKey": "sk-your-api-key",
    "coworkEgressAllowedHosts": ["*"]
  }
}

Do not commit configuration files that contain real API keys. For team deployments, create a dedicated AIOHub API key for Cowork 3P and restrict its groups to the intended use case.

Verify the gateway

First verify that the AIOHub API key and model work:

curl https://api.aiohub.org/v1/messages \
  -H "x-api-key: sk-your-api-key" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4-20250514",
    "max_tokens": 128,
    "messages": [{"role": "user", "content": "Hello from Claude Cowork 3P"}]
  }'

If the request succeeds, launch Claude Desktop and send a short prompt. Then open the AIOHub console Logs page to confirm the request, model, group, and cost.

Troubleshooting

Claude Desktop still opens the normal Claude login

Confirm inferenceProvider is set to gateway, and that inferenceGatewayBaseUrl and inferenceGatewayApiKey are both present. Cowork 3P activates only when the provider and required credentials are valid.

Connection or path errors

Confirm the Gateway base URL is https://api.aiohub.org, without /v1. Cowork 3P appends /v1/messages itself.

Model list is empty

AIOHub supports /v1/models, but visible models still depend on the group assigned to the API key. Usually keep Model list at its default. Only set inferenceModels explicitly, for example claude-sonnet-4-20250514, if the client does not show models correctly.

Authentication fails

Usually keep Gateway auth scheme at its default. If your Claude Desktop build does not infer the sk- API key header correctly, set it explicitly to auto, or try x-api-key.

There is no Chat tab

This is expected in Cowork 3P. The third-party inference mode is centered on the Cowork and Code tabs, and is not the same as standard Claude Chat.

The model cannot access a website

Check whether Allowed egress hosts includes the target hostname. To allow all URLs, set it to *; in JSON configuration, use "coworkEgressAllowedHosts": ["*"].

Official references

• Claude Cowork 3P overview
• Claude Cowork 3P configuration reference
• Install and configure Claude Cowork with third-party platforms