Dify

Dify can call OdiRouter through an OpenAI-API-compatible model provider. Use this for Dify chat apps, agents, and workflows that need OdiRouter LLM models.

Requirements

Requirement	Value
Dify	Cloud or self-hosted workspace with model provider access
Provider type	OpenAI-API-compatible
API endpoint	https://www.odirouter.ai/v1
API Key	OdiRouter API key
Model name	An OdiRouter LLM model ID

Add OdiRouter Provider

1. Open Dify.
2. Go to model provider settings.
3. Add a provider of type OpenAI-API-compatible.
4. Set the API endpoint or Base URL to https://www.odirouter.ai/v1.
5. Paste your OdiRouter API key.
6. Add an OdiRouter model ID, such as gpt-5.5.
7. Save and test the provider.

Use In An App

1. Open a Dify app.
2. Go to model settings.
3. Select the OdiRouter provider.
4. Select the configured model.
5. Run a short test prompt.

Example prompt:

Reply with one sentence confirming that Dify is using OdiRouter.

Read LLM Output

When Dify calls OdiRouter as an LLM provider, the app node usually passes the model text as that node’s output to downstream nodes. If you call chat completions directly from an HTTP request node, the assistant text is at:

choices[0].message.content

In workflows, save this field as a clear variable such as answer, summary, or model_output, then pass it to condition nodes, templates, database writes, or webhook responses.

Multimodal Queue Workflows

For OdiRouter asynchronous image, video, audio, or 3D models, use Dify’s HTTP request capability instead of the OpenAI-compatible LLM provider:

Step	Endpoint
Create task	POST https://www.odirouter.ai/model/v1/queue/{endpoint}
Poll status	GET {status_url}
Fetch result	GET {response_url}

After creating the task, save these fields:

Field	Purpose
request_id	Keep the task ID for logs, debugging, and business records
status_url	Pass to a loop or scheduled HTTP node to poll status
response_url	Fetch the final result after completion

Poll status_url until status is COMPLETED, then call response_url. The final result shape depends on the model type and usually contains image, video, audio, 3D file URLs, or text fields. Map the needed URL or text field into a workflow variable before passing it downstream.

Troubleshooting

Symptom	Fix
Provider test fails with 401	Recreate the OdiRouter API key
Model not found	Add the exact model ID returned by /models
Dify expects a full OpenAI endpoint	Use https://www.odirouter.ai/v1, not /chat/completions, unless Dify explicitly asks for the full chat endpoint
Multimodal model does not work in LLM provider	Use HTTP workflow nodes and the /model/v1/queue/{endpoint} API