Auto: DeepSeek Flash answers normal chats and DeepSeek Pro answers Think harder / Research. Builds, web search, attached files and library documents always use Kimi and are never sent to DeepSeek. If DeepSeek fails, Kimi answers.
Uses your OpenRouter key (below). Cost counts in the monthly limit. Sexual content and fake photos of real people are refused.
All books go into the Library as shared reference with their source and licence. Only books that allow commercial reuse (CC BY, CC BY-SA, CC0, public domain) are used to train Jinnee's own model.
Used in every chat and build, for everyone on your team. Don't put passwords or confidential client data here.
Changes to an existing project are sent as quick edits (only the changed lines). If an edit cannot be placed, Jinnee rewrites the whole file automatically.
Jinnee refuses a request if its worst-case cost would pass either limit. Costs are calculated from the token counts Kimi reports, at list price, so your Kimi bill should be the same or lower.
At least 10 characters. Use a password you don't use anywhere else.
3β40 letters, numbers or hyphens. Anyone with the link can open it.
Published sites show a small "Made with Jinnee Β· Report" badge. Do not publish anything you don't have the right to share.
Only normal chats answered by DeepSeek are used, never chats with attached files or library documents. You can switch this off at any time; your earlier chats are then left out of training too.
Staff can create and edit their own projects. They cannot see the API key, change settings, or see other people's projects. Spending limits are per person per month (blank = no personal limit; the company limit in Settings still applies).