Table of Contents
Overview
Trovata AI Agents extend the capabilities of Trovata AI Chat by turning one-time questions into automated, recurring analyses. Instead of asking the same question every morning, you create an agent once, schedule it, and let it run on your behalf. Agents can run hourly, daily, or weekly.
Each agent operates like a dedicated analyst. It follows the instructions you define, retrieves live data from your Trovata workspace using the same tools available in Trovata AI Chat, and delivers formatted results to an activity feed you can check at any time. You build a fleet of agents, each focused on a specific task: monitoring large transactions, scanning for FX exposure, tracking idle cash, or summarizing your daily cash position, etc.
Agents use the same secure infrastructure as Trovata AI Chat. Your data is never used for model training, never leaves your Trovata instance, and is never accessible to anyone but you.
How Trovata AI Agents Work
When an agent runs, whether on a schedule or triggered manually, it executes a set of instructions you wrote in plain English. Behind the scenes, the agent uses Trovata's verified API tools to retrieve your live financial data, analyzes it, and returns a formatted report.
The agent has access to seven data retrieval tools covering account balances, transactions, cash flow summaries, transaction analysis, account details, payment records, and account analysis. These are the same tools that power Trovata AI Chat. Every number in an agent's output comes from an actual API call to your workspace. Nothing is generated or estimated.
Agents run on AWS Bedrock AgentCore, the same secure, isolated environment used by Trovata AI Chat. The same data privacy guarantees apply: your data is processed securely, never used for training, and never accessible to Anthropic or anyone outside your Trovata instance.
Getting Started
Check Your Access
To use Trovata AI Agents, two things need to be in place:
Feature Enabled: Trovata AI must be turned on for your organization.
Permission Assigned: Your user role needs the Read Agents permission, found under the Trovata AI section in User & Entitlements settings.
If you can't see Agents in your sidebar, contact your organization's Trovata administrator and/or Trovata support (via chat or email at support@trovata.io) to verify that the feature is enabled and your role has the correct permission.
Find Agents in Trovata
Navigate to Trovata AI in the left sidebar, then click Agents. The Agents page opens with two tabs:
Activity is the default view, showing a timeline of all agent execution results grouped by date (Today, Yesterday, and specific past dates).
Listing shows all agents you've created, with their name, status, and last run time.
A New Agent button is visible in the top right corner of both tabs.
Creating Agents
Click New Agent to open the Agent Builder. The builder is a full-screen dialog with two panels:
Left Panel: A chat interface where you describe what you want the agent to do in plain English. A greeting message welcomes you and offers starter templates to get going quickly.
Right Panel: Two tabs. Configuration is where you set the agent's instructions and schedule. Preview is where you test the agent's output before saving.
Describe Your Agent
Type a description of what the agent should do in the chat input. For example:
"Provide a comprehensive overview of all credit transactions from yesterday, including transaction amounts, account details, counterparties, and any relevant descriptive information."
The AI responds conversationally to help you refine the instructions. Your description is automatically populated into the Instructions field on the Configuration tab. You can continue the conversation to iterate, or edit the instructions field directly.
What to include in your instructions: Describe what the agent should monitor, which accounts or data to consider, the timeframe or baseline, how to detect issues (rules or thresholds), and how you want the output formatted (alert, report, key fields). The more specific you are, the more useful the output.
Starter Templates
If you're not sure where to start, the Agent Builder offers five starter templates below the chat greeting:
Daily Cash Position Summary summarizes your cash position across all accounts.
Weekly Cash Flow Report provides a weekly breakdown of credits, debits, and net flow.
Account Reconciliation Monitor checks reconciliation status across your accounts.
90-Day Cash Forecast generates a forward-looking cash forecast.
Large Transaction Alert flags transactions at a threshold you define.
Click any template to send its instructions into the chat. The Configuration tab auto-populates with the template's default instructions and a suggested schedule. You can customize both before saving.
Configuring a Schedule
Switch to the Configuration tab on the right panel. Below the Instructions field, you'll see the Trigger section.
Set the Trigger Type
Click the Trigger dropdown and select Schedule. Three cadence options are available:
Hourly: Select Hourly and choose the minute of each hour when the agent should run (for example, :00, :15, :30, or :45).
Daily: Select Daily and configure:
Time: Choose the hour and toggle AM/PM.
Timezone: Select from a full list of IANA timezones (e.g., America/New_York, America/Los_Angeles, Europe/London).
Multiple Times: Click + Add another time to have the agent run more than once per day. Remove additional entries with the X button.
Weekly: Select Weekly and configure:
Days: Use the multi-select dropdown to choose which days of the week the agent should run (Sunday through Saturday).
Time and Timezone: Same as daily, with support for multiple time entries.
Preview and Save
Preview the output
Before saving, switch to the Preview tab on the right panel. The agent runs a live preview using your current instructions and streams the result in real time. The output may include formatted text, tables, highlighted values, or recommended actions, exactly matching what the agent would produce on its scheduled run.
If you update the instructions and switch back to Preview, a new preview generates automatically with the updated instructions.
Save the Agent
Click the agent name in the top bar (defaults to "Untitled agent") to give it a descriptive name, something like "Daily Credit Transaction Report" or "Weekend Activity Monitor."
When you're ready, click Save. You'll see two options:
Save as draft saves the agent without activating a schedule. The agent appears in your Listing tab with a Draft status badge. It does not run automatically.
Save and activate saves the agent and starts the schedule immediately. The agent appears with an Active status badge and runs automatically at the times you configured.
A confirmation banner ("Agent saved and activated") appears at the bottom of the screen, and the agent is now visible in your Listing tab.
Managing Agents
View Agent Details
Click any agent in the Listing tab to open the detail panel on the right side. The panel shows:
Status: Draft, Active, or other lifecycle states.
Last run: When the agent last executed, or "Never" if it hasn't run yet.
Last modified and Created on dates.
Configuration: The full instructions text and trigger summary (e.g., "Scheduled: Daily").
Execution History: A list of past runs, each with a timestamp, status badge, and a brief summary of results.
Edit an Agent
Open the context menu (three-dot icon) on any agent in the Listing tab and click Edit, or click the Edit button in the detail panel. The Agent Builder opens in edit mode with the existing chat history and configuration loaded. Make your changes and click Save.
Duplicate an Agent
Click Duplicate from the context menu. A copy of the agent is created with the same instructions and trigger settings. The copy is independent, so editing it does not affect the original.
Delete an Agent
Click Delete from the context menu. A confirmation dialog appears. Confirm to permanently remove the agent and its execution history.
Run an Agent Manually
For any active agent, click Execute from the context menu or the play button in the detail panel header. The agent runs immediately regardless of its schedule, and the result appears in the Activity feed within seconds to a minute.
Monitoring Agent Activity
The Activity tab is your dashboard for all agent executions. Results are grouped by date, with the most recent at the top.
Each entry shows the agent name, a brief summary of the result, a timestamp, and a color-coded status badge:
Success (green) indicates the agent found actionable results.
No Action (gray) indicates the agent ran but had nothing to report.
Warning (yellow) indicates a condition worth your attention.
Alert (red) indicates a critical condition was detected.
Error (red) indicates the execution failed.
Click any entry to open the full output in a detail panel. Results may include formatted text, data tables, highlighted values, and recommended actions.
Use the search bar on either tab to filter agents or activity entries by keyword.
Available Data Sources
Agents have access to seven data retrieval tools. Each connects to a specific part of your Trovata workspace:
Account balances provide current and historical balances, grouped by institution, entity, currency, or account type. Supports daily, weekly, monthly, and quarterly cadences.
Account details return individual account records with balances, connection info, and account metadata.
Transactions return individual transaction records with full details. Filter by date, amount, type (credit/debit), account, tag, keyword search, and more.
Transaction summaries provide aggregated transaction totals and counts, useful for high-level reporting without individual transaction detail.
Transaction analysis shows transaction trends over time, segmented by tag, institution, currency, entity, or other dimensions.
Account summaries deliver a high-level cash position overview across accounts, grouped by any account attribute.
Payments return payment records filtered by status (open, submitted, approved, failed, completed, and others) and type (ACH, Wire, RTP, Transfer).
All tools accept optional filters. If no filters are provided, sensible defaults are applied (e.g., date range defaults to the last 90 days, currency defaults to USD).
Tips & Tricks
The quality of an agent's output depends on the instructions you give it. A few principles that make a difference:
Be specific about scope. "Monitor all accounts" is broad. "Monitor my USD accounts at Chase and Bank of America" gives the agent a clear target.
Define the timeframe. "Check yesterday's transactions" or "analyze the last 7 days" is precise. "Check recent activity" is ambiguous.
Set thresholds when relevant. "Flag any single transaction over $100,000" gives the agent a clear rule. "Flag large transactions" leaves too much interpretation.
Describe the output format. "Provide a summary table with columns for amount, account, counterparty, and description" tells the agent exactly how to structure the result.
Iterate using chat. The Agent Builder's chat interface is there to help you refine. Start with a rough idea, review the AI's response, and tighten the instructions based on what you see.
FAQs
General
What's the difference between Trovata AI Chat and Agents? Trovata AI Chat is for interactive, one-time questions where you ask and get an answer in real time. Agents automate those same kinds of questions on a schedule. Think of Chat as asking a question and Agents as assigning a recurring task.
Can I have multiple agents running at the same time? Yes. You can create as many agents as you need, each with its own instructions and schedule. The Activity tab gives you a unified timeline across all of them.
Do agents share context with each other or with Trovata AI Chat? No. Each agent operates as an independent session. Agents do not share context with each other or with your Trovata AI Chat conversations.
Security
Are agents subject to the same security protections as Trovata AI Chat? Yes. Agents run on the same AWS Bedrock AgentCore infrastructure. Your data is never used for model training, never stored beyond execution, and never accessible to Anthropic or anyone outside your Trovata instance.
Can other users in my organization see my agents? Agent visibility follows your organization's permission model. Users need the Read Agents permission to access the Agents feature.
Creating and Configuring
Do I need to write code or know an API to create an agent? No. You describe what the agent should do in plain English using the chat interface. The AI helps you refine the instructions. No coding, query language, or API knowledge required.
Can I edit an agent after saving it? Yes. Open the agent from the Listing tab, click Edit, and modify the instructions or schedule. Click Save to apply your changes.
What happens if I save without configuring a schedule? The agent is saved as a Draft. It appears in your Listing tab but does not run automatically. You can add a schedule later by editing the agent.
Execution and Results
How long does an agent take to run? Most agents complete within seconds to a minute, depending on the complexity of the instructions and the volume of data being retrieved.
What happens if an agent fails? The execution shows an Error status badge in the Activity feed. You can review the error, adjust your instructions if needed, and re-run the agent manually.
Can I run an agent manually outside its schedule? Yes. Use the Execute option from the agent's context menu or detail panel to trigger an immediate run at any time.
Troubleshooting
I can't see Agents in my sidebar. Two things need to be in place: Trovata AI must be enabled for your organization, and your user role needs the Read Agents permission under the Trovata AI section in User & Entitlements. Check with your Trovata administrator.
My agent's output doesn't match what I expected. Review the instructions on the Configuration tab. Vague or broad instructions produce general results. Add specificity: name accounts, set date ranges, define thresholds, and describe your preferred output format. Use the Preview tab to test changes before saving.
An agent shows "Error" in the Activity feed. This can happen when data is unavailable, a date range is invalid, or there's a temporary service issue. Try editing the agent to adjust the instructions or time window. If errors persist, contact Trovata support.