Choose models per chat with less setup
Why choose a model for a group?
If you start several chats for translation or summaries, repeating the same model and instructions in every new chat takes time. Set them for the group once, then change only the chats that need a different model or instruction. A change to a shared Global value reaches chats that inherit that value; the settings are used from their source, not copied into each chat.
Which model answers this message?
If this Chat has a model, it uses that model. Otherwise, it checks its group, then the Global model; a Chat outside a group goes straight to Global. The first place with a model decides which one answers. Using a setting from a broader scope this way is called inheritance. The connection is chosen separately in the same chat → group → Global order.
Each scene sends the same question independently. In this example, all three models are available from the same Ollama server; only the model setting changes.
The arrows show the order used to find a model setting. Once Naidan finds a model, it uses that model and stops looking higher up. The connection is chosen separately in the same order.
Set these values in Naidan
First, open Naidan. If first-run setup appears, close it with the × at the top right, then follow step 1.
The addresses and model names below illustrate the controls. Select models actually available from your own servers.
1. Set the global starting point
Open Settings at the bottom of the sidebar, then Connection. Choose an API Provider and Endpoint URL, select a Default Model, and save. This example uses Ollama. These values supply the fallback for chats without a more specific setting.
2. Set a group model and, if needed, its endpoint
If you do not yet have a group with a Chat in it, follow Group chats and settings by purpose to create a group and add a Chat before continuing.
Select a chat group’s name in the sidebar to open its settings page. For a different model on the same server, leave Endpoint Type inherited and select a model in Model ID Override. The figure also selects Ollama as a group endpoint override: this copies the current Global URL and HTTP headers, so the URL stays the same unless you edit it. Later Global URL and header changes do not reach that group override.
3. Read what a chat inherits
Open a chat in the group and select the model name and gear in its header. In the chat settings dialog, the Endpoint Type choice and the empty URL and Model Override controls show the effective group values as inherited labels. Empty local controls still use those values.
4. Change only this chat’s model
Choose a different available model in Model Override. It becomes this chat’s local model, while its endpoint still comes from the group. Chat changes save from this dialog without a separate global Save Changes button.
5. Change a chat endpoint, then inherit again
A local Endpoint Type or URL creates a complete chat endpoint override. If that change points to a different model namespace, Naidan clears this chat’s local model choice; select a model from the new endpoint. To return to the group setting, choose the parent value in Endpoint Type. A namespace change also clears a local model choice; if a model override remains, choose the inherited entry at the top of Model Override. Clearing the URL text alone does not remove an endpoint override.
To restore only the connection or model, choose its inherited entry above. The chat or group Restore defaults action also resets other overrides in that settings panel. Automatic title generation has separate settings, so changing the normal chat endpoint or model does not by itself replace an inherited title-generation target.