syntaqxThe top bar's Ask is live: it opens a conversation with g1t in a panel beside the page, answered by the AI provider and model the workspace's owner chose, streamed as it is written, with Markdown drawn as chat draws it, the provider and model under each answer, Stop to end an answer (what came stays), and New conversation to start over. Each question carries the page you are on, so g1t can say which repository, issue or pull request you are looking at. The conversation is kept in the browser tab for now, across pages and reloads; closing the panel keeps an answer coming. Until a provider is chosen the panel says so: owners get a link to choose one, members are told to ask an owner. On Integrations, Your AI providers has an Ask g1t card where owners choose the provider and one of the models it offers (a deployment for Azure, any name for an endpoint that lists none), or turn Ask off, and members see what it answers with. Questions go through -/ask, which streams the integrations service's answer straight through and counts 20 questions a minute per person (WEB_ASK_LIMIT).763e965History
