AI Agents
Understand and configure AI agents — autonomous assistants that analyze data, write reports, detect anomalies, and more on your behalf.
What Are AI Agents?
AI agents in OpenAnalyst are autonomous task executors that combine access to your data, a selected AI model, and a defined skill set to accomplish analytical goals with minimal manual input. Unlike a simple query or chart, an agent can plan multi-step tasks, call tools, reason about intermediate results, and produce structured outputs such as reports, dashboards, or recommendations.
Agents operate within your workspace context — they can access the connectors, dashboards, and reports you have permission to see, and they respect all existing data governance rules.
Available Agent Types
| Agent | Primary Capability | Best For |
|---|---|---|
| Data Analyst | Explores datasets, surfaces insights, answers ad-hoc questions. | Exploratory analysis, executive summaries. |
| Report Writer | Generates structured narrative reports from data with charts and commentary. | Weekly business reviews, board decks. |
| Anomaly Detector | Scans datasets for statistical outliers and unusual patterns. | Fraud detection, infrastructure monitoring. |
| Forecaster | Builds time-series forecasts using statistical and ML models. | Revenue forecasting, demand planning. |
| SQL Expert | Generates, optimizes, and explains complex SQL queries. | Query optimization, schema documentation. |
| Data Cleaner | Identifies and corrects data quality issues — duplicates, nulls, type mismatches. | Data preparation before analysis or migration. |
Configuring an Agent
Each agent can be configured with a name, a base AI model, a set of enabled skills, and context such as which data sources it should prioritize. Navigate to Settings > AI Agents and click New Agent.
- Select an agent type from the list above.
- Choose the AI model to power the agent (see the AI Models page for guidance on which model to select).
- Enable or disable individual skills — each agent type ships with a default skill set that you can customize.
- Set data source scope: all connectors, specific connectors, or specific schemas.
- Optionally add a system prompt to shape the agent's tone, focus area, or output format.
- Save the agent. It is now available in the agent panel and via the API.
Agent Conversations
Interact with any agent through the conversational interface accessible from the right-hand panel on any page. The conversation is threaded — follow-up questions retain context from earlier messages in the same session. You can reference charts, tables, or data sources by name and the agent will resolve them automatically.
You: Summarize the key trends in our Q1 sales data.
Agent: Based on your Q1 Sales dashboard, here are the three key trends:
1. Total revenue grew 14% month-over-month in March, led by the Enterprise segment.
2. The APAC region showed the highest growth rate at 32%, though it remains the smallest by volume.
3. Churn-adjusted net revenue retention improved from 108% in January to 117% in March.
Would you like me to generate a full report or drill into any of these areas?Agent Memory and Chaining
Agents maintain a session memory that persists for the duration of a conversation, allowing them to reference earlier results, build on previous queries, and avoid redundant data fetches. Memory is scoped to the session and is not shared between separate conversation threads.
Agent chaining allows the output of one agent to become the input of another. For example, the Data Cleaner agent can preprocess a dataset and pass the cleaned result to the Forecaster agent, which then generates predictions. Chains are configured in the agent editor under the Pipeline tab, where you connect agents in sequence and map output fields to input parameters.
Tip: Use agent chains for recurring workflows. Save a chain as a scheduled pipeline to run automatically on a daily or weekly basis, delivering results to a specified dashboard or Slack channel.