NarraLume

Xudeng is a local priority AI novel writing studio. It puts creative intentions, outlines, characters and settings, timelines, text versions, review opinions, and AI collaboration tasks into the same library, helping you write from an idea to a deliverable manuscript. ##Model Configuration Before using AI for the first time, open "Settings" on the left and configure it strictly in the order of "Channel → Model → Position" in "Channel and Model Management". When creating channels only, it is normal for the page to display "0 model"; it is also necessary to manually register the model and assign the position. ###Step 1: Create a new model channel Click "New Channel" and fill in the following: | field | completion instructions | | ---------------- | -------------------------------------------------------------------------------------------- | | channel name | Used only for identification, such as "DeepSeek","OpenAI" or the name of the compatible service you use. | | agreement | Must be consistent with the service provider's actual API: OpenAI Chat Completions, OpenAI Responses, or Anthropic Messages. | | Base URL | The API root address provided by the service provider. Don't write API keys into URLs. | | Endpoint (optional)| Normally left blank; only filled in if the service provider clearly specifies a non-standard request path. | | key | Fill in the service provider API Key. The input box will be hidden. | | enabled | Keep checking, otherwise models under this channel cannot be assigned. | After saving, use the "Probe" or connection test in the channel card to confirm that it is available, and then create a new model. ###Step 2: Create a new model under the channel Select the channel you just saved and click "New Model" on the right. | field | completion instructions | | ---------- | -------------------------------------------------------------- | | Upstream model name| The model ID actually accepted by the service provider is not the display name created by itself. | | model type | Choose "writing" for text creation. The embedding model is only used for semantic retrieval and does not participate in body generation. | | Context upper limit| Fill in when you know the service provider's upper limit; leave it blank if you are not sure. | | output upper limit | It is recommended to fill in according to the service provider's restrictions; you can leave it blank if you are not sure. | | enabled | Keep checking. | After saving the model, the channel card no longer displays "0 Model". ###Step 3: Assign positions Select the writing model you just created in "Default Generation Model". By default, it is used for writing, planning, and review; only when different capabilities or cost strategies are really needed, the planning or review model is covered separately. After completing the assignment, go back to the bookshelf to create a blank work, first write a short paragraph of text by hand, and then initiate a minimum AI editing or continuation request. This distinguishes model connectivity issues from work data issues.