> ## Documentation Index > Fetch the complete documentation index at: https://docs.rewards.so/llms.txt > Use this file to discover all available pages before exploring further. # Assistant > Rewards AI: ask questions about your program in plain language, and make changes without leaving the page. Rewards AI is the panel behind the bubble at the bottom right of the dashboard. Ask it something in your own words and it looks it up, or prepares the change for you to approve. It runs on exactly the same tools as the [MCP server](/app/mcp), so anything an external AI client can do with your program, the assistant can do too. *** ## What you can ask it It covers your loyalty program and the product itself. **About your data** ``` Who is at the top of the leaderboard? How did the last 30 days go compared with the month before? How many points does 0x1234... have? Which automations are active? Who changed the point triggers last week? ``` **Making changes** ``` Add 500 points to vitalik.eth for the discord_launch event Create a trigger connect_telegram worth 50 points Prepare a draft campaign for the top 100, 0.001 SOMI each Ban 0xdead for spam ``` **About Rewards itself** ``` How do I award points from my backend? What does the /leaderboard endpoint return? How do reward automations decide who gets in? ``` For product questions it answers from this documentation rather than from memory, and links to the page it used. *** ## Changes always need your approval The assistant never changes anything on its own. When a request would add points, ban a wallet, create an automation or edit a draft, it stops and shows you exactly what it is about to do. Nothing runs until you press **Confirm**. This is enforced by Rewards, not by asking the assistant nicely: a change it has not had approved cannot execute. Your role decides what it will offer. Owners and admins can make changes; moderators and viewers get the same answers but no write actions, on the programs where they hold that role. *** ## What it will never do It cannot publish, cancel or distribute a campaign on-chain. Those need a signature from your wallet and only exist in the dashboard. The furthest it goes is preparing a **draft**: it can build a campaign down to the last recipient, then tell you it is ready for you to publish yourself. *** ## Working across programs If you have several point systems, the assistant starts across all of them. Once the conversation settles on one it stays there, and the header shows which. Ask about another by name and it will go and look, coming back to the general view when an answer spans more than one. You can also pick a program yourself from the header. Doing so starts a new conversation, since wallets and campaigns from one program mean nothing in another. *** ## Conversations Conversations are kept for **15 days**, tied to your account, so you find them again on any device you sign in from. After 15 days without activity a conversation is deleted automatically. You can delete one yourself at any time from the list. The bar under the header names the current conversation after your first question, lists the others, and holds the buttons to start one or delete one. Two actions sit in the header once a conversation has started: | | | | ----------------- | ----------------------------------------------------------------------------------------------------- | | **Download** | Saves the conversation as a Markdown file, including the actions that ran | | **Email support** | Saves that file and opens a message to support with the context filled in, ready for you to attach it | *** ## Language The assistant replies in the language you picked for the dashboard, whichever language you type in. *** ## If it cannot answer It will say why rather than failing silently: a provider limit reached, an account out of credit, a key problem. If something looks wrong, download the conversation and send it to support: the file records what ran and what came back, which is usually enough to explain it.