Use model chat
Open your session from Sessions, then select Open model chat once it is available. If you are returned to the session page, check contributor readiness.
Send a message¶
Type in Your message.
Select Send message, or press Enter.
Wait for the reply to finish before sending another message.
Use Shift+Enter to insert a new line. Replies appear as the model generates them. The Copy button beside a message copies its text.
Follow-up messages continue the conversation. Select New chat to clear the visible conversation and start fresh. Save any replies you want to keep before doing so; do not rely on the chat page as a permanent conversation archive.
Reply length and context¶
The session page’s API access section shows the Reply cap: the maximum number of output tokens per reply. A token is a piece of text, not necessarily a whole word. A reply can end because it reaches that cap.
The model also has a context limit for the conversation and its answer. If the conversation becomes too long, start a New chat and include a shorter summary of what the model needs to know.
Optional direct chat¶
If your session supports it, open Settings and select Enable direct chat. This tries a connection to the first and last contributors, with coordinator fallback. Direct chat can expose your public IP address to those contributors; Private TURN sessions use TURN instead.
Wait for connection setup to finish before sending. Live chat connections shows whether traffic uses direct links, TURN or coordinator fallback. The fallback is still a working chat path; direct chat is optional.
If a reply stops¶
Check the displayed error and return to the session page to inspect contributors. A disconnected contributor can interrupt inference. After availability returns, retry manually or select New chat. Avoid sending repeated requests while the session is unavailable.
For an application instead of the chat page, see Use the API.