Skip to content

Your First Chat

Start with a running Numi instance and an available model. If you are setting up your own server, complete Getting Started first.

Send a question, receive a reply, and file the conversation to continue later.

Sign in

Open your instance's address. For a local installation, use http://localhost:4567/chat.

If you do not have an account, choose Sign up on the sign-in page. Enter a username and a password that meets the displayed requirements, then select Sign Up. Otherwise, sign in with your existing account.

Send a message

  1. Select New Chat in the sidebar. On a narrow screen, open the sidebar first.
  2. Check the model selector beside the message input. Keep the selected model for this walkthrough, or choose another available model.
  3. Enter this request and press Enter or select the send button:
I have 30 minutes to tidy my desk and sort my notes. Suggest a three-step plan.

The answer appears as it streams. Some models also return a reasoning trace, shown in a collapsible block. A trace is not guaranteed: it depends on the model, provider, and reasoning setting. See Reasoning and its limits for how stored reasoning is reused.

Use Shift+Enter to add a line break without sending.

Continue and reopen the conversation

In the same chat, send:

I only have 10 minutes now. Shorten that plan.

The response should use the task you already described. This checks conversation continuity; it does not prove that the provider replayed a reasoning trace.

After the answer finishes, refresh the page and select the conversation in the sidebar. Both messages and their replies should still be there. You now have a saved conversation you can continue later.

To try a task that uses external data, continue with Working with Tools. To keep this chat easy to find, pin it or search its messages.

If the first response fails

  • The site does not open: check that the server is running and that you are using its address. localhost refers to the device running your browser.
  • No model is available: the instance operator needs to configure a provider and enable a model.
  • A provider error appears: retry after the connection recovers, or choose another available model. The operator can check the server logs and provider configuration if it persists.