Your First Chat¶
Start with a running Numi instance and at least one configured model. If you are setting up a new server, finish Install Numi first.
Sign in¶
Open your instance URL in the browser (http://localhost:4567/chat for a local
install) and sign in. If this is a fresh database, choose Sign up to create
the initial administrator account.
Send a message¶
- Click New Chat in the sidebar (on mobile or narrow windows, open the sidebar menu first).
- Check the model selector next to the composer. Keep the default model or pick another available model. To start every new chat with your own choice, set Settings -> General -> Default model; it is saved with your account and applies on all your devices.
- Type a prompt and press Enter (or click the send button):
The response streams in token by token. When the selected model returns reasoning tokens, a collapsible thinking block appears above the reply so you can inspect its intermediate steps. See How reasoning is preserved for how stored reasoning is handled on later turns.
In Settings → Appearance, choose Light and the Claude color palette for a white chat with serif replies. Reading font overrides the reply font without changing the sidebar, message box, or code.
Useful composer and message controls¶
- Line breaks: Press Shift+Enter to insert a newline without sending.
- Send while Numi is still working: Press Enter during a streaming turn to queue your follow-up for after the current answer finishes, or press Ctrl/Cmd+Enter to inject it right after the current tool batch completes. See Sending a message while Numi works.
- Regenerate and compare versions: Click Regenerate response below an
assistant reply to try again (even with a different model). Use the version
arrows (
< 1 / 2 >) to flip between earlier and newer replies. Editing one of your own messages or its attachments also creates a selectable branch without erasing the original thread. - Temporary chats: Toggle Temporary at the top right of a new chat when you want an off-the-record conversation. Temporary chats never write to long-term memory and are automatically deleted after you leave or after 24 hours of inactivity. To turn off memory across your whole account, toggle Use memory in Settings -> Privacy & data.
Ask a follow-up¶
In the same chat, send:
Numi sends the conversation history (and, on supported models, the prior turn's reasoning trace) so the model can revise the plan directly in context. Refreshing the browser or reopening the chat from the sidebar restores the full thread.
Next, try Working with Tools to run Python calculations, search the web, and use Deep Research, or read Chats, search, and sharing.
Troubleshooting¶
- The page does not load: Confirm the server process or Docker container is
running and that you are opening the right host and port (
localhostalways points to the device running your browser). - No model is available: An administrator needs to configure a provider and enable at least one model on the Models page.
- Provider error mid-turn: Click Regenerate response or switch to another model in the selector. If the error persists, check the server logs and provider API key.