> ## Documentation Index > Fetch the complete documentation index at: https://conto.finance/docs/llms.txt > Use this file to discover all available pages before exploring further. # AI Assistant > Natural language interface for managing agents, wallets, policies, and more # AI Assistant The AI Assistant (Conto AI) is a conversational assistant built into the dashboard that lets you manage your entire agent payment infrastructure using natural language. Instead of navigating forms and tables, you can type what you want and the assistant handles the rest. ## AI Assistant vs Conto Pay vs MCP Conto has three natural-language surfaces that differ by who acts, how you reach them, and whether they move funds onchain. The AI Assistant is the one where **you** act, from the dashboard chat panel: it configures agents, wallets, and policies, but does not execute payments onchain. The other two are [Conto Pay](/conto-pay/overview) (a Conto-hosted payment agent) and the [MCP Server](/mcp/overview) (your own external AI agent authenticated with an agent SDK key). See the [three-surface comparison table](/mcp/overview#mcp-vs-ai-assistant-vs-conto-pay) for the full breakdown. ## Opening the Assistant There are three ways to open the assistant panel: * **Floating button**: Click the purple sparkle button in the bottom-right corner of any dashboard page * **Sidebar link**: Click "Assistant" in the left navigation sidebar * **Keyboard shortcut**: Press `Cmd+J` (Mac) or `Ctrl+J` (Windows/Linux) from any page The assistant opens as a slide-out panel on the right side. You can continue using the dashboard while the panel is open. ## What You Can Do The assistant registry currently exposes **82 tools** across **8 categories**. That includes the core dashboard admin and read tools plus **39 hosted Conto Pay tools** that appear when the assistant is operating a Conto Pay workspace. ### Query Data Ask questions about your organization's state: ``` "How many agents do I have?" "Show me all transactions over $500 this week" "Are there any critical alerts?" "What's the trust score for our OpenAI counterparty?" "Give me a spending overview for all agents" "Explain what the overnight-guardrail policy actually does" ``` ### Manage Agents Create, configure, and control agents: ``` "Create an agent called Payment Bot of type OpenAI Assistant" "Pause the Marketing Agent" "Assign the spending-limit policy to the DevOps agent" "Generate an SDK key for the Research Agent" "Link the Operations wallet to Payment Bot with a $500 daily limit" ``` ### Configure Wallets Create wallets and manage spend limits: ``` "Create a new EVM wallet called Treasury" "Set the daily spend limit to $1000 for Payment Bot's wallet" "Freeze the compromised wallet" "What's the balance on all my wallets?" ``` ### Create Policies Describe rules in plain English and the assistant translates them into the correct policy type and rules: ``` "Create a policy that blocks spending over $500 per day" "Make a rule that only allows transactions Monday through Friday 9am-5pm EST" "Block all transactions to addresses not on the whitelist" "Require manual approval for any transaction over $1000" "Create a policy that denies gambling category transactions" ``` The assistant infers the correct `policyType`, `ruleType`, `operator`, and `action` from your description. If your description is ambiguous, it will ask for clarification. ### Manage Counterparties Track vendors and manage trust: ``` "Add a new counterparty called Stripe with type VENDOR" "Block the suspicious counterparty" "Verify the AWS counterparty" "Recalculate trust scores for all counterparties" ``` ### Handle Approvals Review and act on pending approval requests: ``` "Show me all pending approval requests" "Approve request clxyz123, looks legitimate" "Deny that request, the amount is too high" "Create an approval workflow that requires 2 approvals for transactions over $5000" ``` ### Bulk Operations Act on multiple entities at once: ``` "Pause all agents that have exceeded their daily limit" "Assign the new security policy to all active agents" "Set a $200 per-transaction limit for all agents" ``` ### Operate Conto Pay When the assistant is working inside a hosted Conto Pay workspace, it follows the normal hosted payment lifecycle: draft, authorize, approve when needed, then send. ``` "Show my Conto Pay workspace and payment draft" "Create a Conto Pay request link for 250 pathUSD with memo Invoice INV-100" "Search Conto Pay for Vendor Org and add it as a network recipient" "Approve the latest Conto Pay payment" "Send the approved Conto Pay payment now" ``` ### Detect Anomalies Proactive risk analysis: ``` "Are there any unusual spending patterns?" "Which counterparties have low trust scores?" "Show me all failed transactions this week" ``` ## Confirmation for Dangerous Actions The assistant won't silently execute high-impact actions. Depending on the risk level, you'll see an inline confirmation card before the action proceeds. ### Destructive Actions (Red) These require explicit confirmation with details about the impact: * Deleting an agent or policy * All bulk operations (pause all agents, assign policy to all, etc.) ### High-Impact Actions (Amber) These show a confirmation with a single "Confirm" button: * Suspending an agent * Freezing a wallet * Blocking a counterparty * Approving or denying payment requests * Generating SDK keys * Unlinking wallets or removing policies ### Preview Actions (Green) Creation tools show a preview card of what will be created, informational, not blocking: * Creating agents, wallets, policies, counterparties * Linking wallets to agents * Assigning policies * Updating spend limits (shows before/after) ## How It Works The assistant turns your request into the right Conto actions and streams the results back in the UI. When you send a message: 1. Your message is sent to the streaming endpoint 2. The assistant analyzes your intent and selects the appropriate action(s) 3. Requests run against your organization's data with normal permission checks 4. Results stream back in real-time with progress indicators 5. The assistant summarizes the results in natural language All assistant actions are scoped to your organization and recorded in the audit log. ## Limitations * **Authentication required**: The assistant only works for logged-in dashboard users. It cannot be accessed via API or SDK. * **Organization-scoped**: All operations are restricted to your current organization. You cannot query or modify data in other organizations. * **No raw blockchain execution**: The dashboard assistant can create wallets and manage configurations, but it does not expose arbitrary direct onchain sends. Hosted Conto Pay payment tools still run through the normal draft, authorization, approval, and send flow inside Conto Pay. * **Rate limited**: 30 messages per minute per user. If you hit the limit, wait a moment and retry. * **Message length**: Individual messages are limited to 4,096 characters. * **Session persistence**: Conversation history is saved in your browser and persists across page navigations and tab closures. Clearing browser data clears the history. * **No file uploads**: The assistant works with text only. You cannot upload CSVs or documents. * **Policy inference**: While the assistant is good at translating natural language to policies, complex multi-rule policies may need adjustment. Always review created policies in the Policies page. ## Tips * **Be specific**: "Create an agent called Research Bot of type Anthropic Claude" works better than "make an agent." * **Reference by name**: "Pause the Marketing Agent" is clearer than "pause agent clxyz123." The assistant searches by name. * **Ask for help**: "What can you do?" or "How do I set up a spending limit?" and the assistant will guide you. * **Review after bulk ops**: After bulk operations, check the affected entities in the dashboard to verify the changes. * **Use Cmd+J**: The keyboard shortcut is the fastest way to toggle the assistant from any page.