> ## Documentation Index > Fetch the complete documentation index at: https://docs.adriel.com/llms.txt > Use this file to discover all available pages before exploring further. # AI Agent > Conversational assistant for interactive analysis and questions about connected data. The **AI Agent** is Adriel's conversational assistant. Ask a natural-language question — "which campaign drove the most conversions last week?", "why did CTR drop on Meta?", "compare organic vs paid reach" — and the agent researches, analyzes, and answers, drawing on every connected data source in the workspace. Use the AI Agent for interactive analysis: iterating on questions, pulling numbers on demand, or getting a second look at what the data shows. ## Open the AI Agent The AI Agent can be opened three ways: * **Main left navigation** — opens the AI Agent as a **full-page** chat experience. Best for extended conversations or working in the agent without a dashboard in view. * **Dashboard top header** — opens the AI Agent as a **side panel** on the current dashboard, so the dashboard stays visible alongside the conversation. * **AI button on a widget** — opens the AI Agent as a **side panel** with that widget already attached as context. The full-page and side-panel views share the same conversations — a chat started in one appears in the other. ## Start a conversation The starting page shows a set of preset prompts to give ideas. Click one to use it, or type any question into the input box. Click **Add Context** below the input box to give the agent scope. Attach a teamspace, a dashboard, or specific widgets to focus the answer on that data. Depending on how the AI Agent was opened, context may already be attached. Opening from the dashboard top header pre-attaches the current dashboard; opening from a widget's AI button pre-attaches that widget. Full-page access from the main navigation opens with no context, and the agent may ask clarifying questions until context is added. Press Enter or click the send button. The agent processes the question and streams a response back. Give any response a thumbs up or thumbs down for feedback. Follow up with more questions in the same conversation. The agent keeps context across turns. ## Manage conversations Where past conversations appear depends on the view: * **Full-page view** — conversations show up in the sidebar on the left. Click any past conversation to reopen it and continue. * **Side-panel view** — conversations show up as a dropdown at the top of the panel header. Pick any past conversation from the dropdown to reopen it. Each conversation is **user-based** — other members of the workspace can't see your chat history. The only exception is **Admin** members, who can see every conversation in the workspace. ## Scheduled prompts The AI Agent can run prompts on a recurring schedule and deliver the results without a user opening the chat. Use scheduled prompts for weekly recap analyses, daily anomaly checks, or any question that would otherwise need to be re-asked periodically. **Scheduled prompts are created inside a chat, in natural language.** Ask the agent to "run this analysis every Monday morning" or "check my Meta CPA every day and let me know if it exceeds \$50" — the agent recognizes the scheduling request and confirms the schedule before saving. **The scheduled prompt runs in the same conversation it was created in.** Each scheduled run appends the new prompt and the agent's response to the original conversation thread — no separate thread is created. This means the full history of scheduled runs lives alongside the original conversation, so you can scroll through past results in one place. When a scheduled run completes and you haven't opened the chat yet, the conversation is marked as **unread** and a **notification counter** appears on the chat — in the conversation list (full-page view) or on the chat entry in the side-panel dropdown. Opening the conversation clears the counter. Once created, scheduled prompts live in the **AI Settings** modal under **Scheduled prompts**. From there: * View every scheduled prompt in the workspace. * Pause, resume, or delete an existing scheduled prompt. * Review the prompt text and cadence. The AI Settings modal is where scheduled prompts are **stored and managed**. New scheduled prompts are only created from inside a chat conversation. ## Custom instructions The **AI Settings** button in the AI Agent header opens the settings modal, which also holds **Custom instructions**. Custom instructions are a rich-text block that tells the AI Agent about the business, goals, tone preferences, or reporting standards — the agent uses these instructions in every response so they don't need to be re-explained in each conversation. Custom instructions can be set at two scopes, chosen from the scope selector at the top of the Custom instructions tab: * **Workspace** — instructions that apply to every conversation in the workspace, regardless of which teamspace is in context. * **Per teamspace** — instructions scoped to a specific teamspace. Any conversation that has that teamspace attached as context picks up the teamspace-level instructions on top of the workspace-level ones. When both scopes have instructions and both apply to a conversation, **teamspace instructions take precedence** on any conflict. If only one scope has instructions, only that one is applied. Teamspace-level custom instructions are useful when different teamspaces represent different clients, products, or reporting audiences. For example, a **Client A teamspace** might carry instructions like "the client cares primarily about ROAS; always call out any campaign whose ROAS falls below 2." A **Client B teamspace** could carry a different priority — "focus on brand-safety metrics; flag any impressions on unapproved placements." Users switching between the two get responses tailored to each client's reporting priorities without re-typing the context each time. ## Permissions * **View only** members can open the agent and view past conversations they have access to. * **View and edit** members can send new prompts and manage their own conversations. ## Troubleshooting The agent needs a teamspace, dashboard, or widget selection to answer accurately. Click **Add Context** and attach the relevant scope. Give the response a thumbs down and add feedback. Retry with a more specific prompt, or add context to narrow the scope. Confirm the prompt is not paused in AI Settings → Scheduled prompts. Confirm the schedule cadence matches what's expected. Widget creation via the AI Agent is not currently supported. The feature existed previously but was removed. The agent returns a message indicating that widget creation is unsupported; if it appears stuck instead, that indicates a downstream error rather than the feature working — contact support with the prompt and a timestamp. ## Related Rule-based monitoring for marketing data and connectivity. Workspace-level scheduled CSV and GSS exports.