Agent Catalog
The Agent Catalog is the central registry of all agents available in your workspace. It includes pre-built agents provided by Datafi, agents shared by your organization, and custom agents you have created. From the catalog, you can browse by category, inspect agent specifications, run agents on demand, and review results.
Browsing the Catalog
Navigate to AI > Agent Catalog to open the catalog view. Agents are displayed as cards, each showing the agent name, description, version, and category.
Search and Filter
- Search -- Type keywords in the search bar to find agents by name or description.
- Category filter -- Use the sidebar filter to narrow agents by category.
- Tags -- Agents can be tagged with custom labels. Filter by tags to find agents related to a specific domain, data source, or use case.
- Sort -- Sort by name, most recent, or most used.
Categories
Agents are organized into functional categories:
| Category | Description | Examples |
|---|---|---|
| Data Analysis | Agents that query, aggregate, and summarize data. | Revenue summary, churn analysis, inventory audit |
| Data Quality | Agents that validate, profile, and clean data. | Null checker, schema drift detector, duplicate finder |
| Reporting | Agents that generate formatted reports and dashboards. | Weekly KPI report, compliance summary, executive brief |
| Document Processing | Agents that extract and transform document content. | Invoice extractor, contract reviewer, receipt parser |
| Integration | Agents that move data between systems via APIs and connectors. | CRM sync, ERP loader, webhook dispatcher |
| Monitoring | Agents that watch for anomalies, thresholds, or events. | Cost spike alert, SLA monitor, pipeline health check |
Agent Details
Click any agent card to open the detail view. The detail view shows the full agent specification, organized into tabs:
Overview Tab
- Name and version -- The agent's identity and current version number.
- Description -- A human-readable summary of what the agent does.
- Goals -- The specific objectives the agent is designed to achieve.
- Success criteria -- Measurable conditions that define a successful run.
- Author -- The user or team that created the agent.
- Last updated -- When the agent was most recently modified.
Capabilities Tab
Lists the tools, data sources, and output formats the agent uses. See Agent Builder for a full reference of available tools.
Behavior Tab
Shows the agent's execution mode, reasoning strategy, retry policy, and memory configuration.
Guard Rails Tab
Displays the constraints and resource limits applied to the agent, including token budgets, API call limits, PII filtering rules, and approval requirements.
Running an Agent
On-Demand Execution
- Open the agent detail view.
- Click Run Agent.
- Provide any required input parameters (displayed as a form based on the agent's input schema).
- Click Execute.
- Monitor progress in the execution panel. You see each step the agent takes in real time.
- Review results when the run completes.
Triggered Execution
Agents can also run automatically based on triggers. If an agent has triggers configured, the detail view shows them in the Triggers section:
| Trigger Type | Description |
|---|---|
| Manual | Run on demand from the catalog or API. |
| Schedule | Runs on a cron schedule (e.g., every Monday at 8 AM). |
| Event | Fires when a platform event occurs (e.g., new data loaded, policy violation detected). |
| Webhook | Fires when an external HTTP request hits the agent's webhook URL. |
| Polling | Checks a condition at a defined interval and runs when the condition is met. |
Triggers are configured in the Agent Builder. The catalog displays triggers as read-only information.
Result Types
When an agent completes a run, it produces results in one or more output formats. The catalog renders each format appropriately in the results panel.
| Result Type | Rendering | Use Case |
|---|---|---|
| Text | Plain text block in the results panel. | Summaries, explanations, status messages. |
| JSON | Syntax-highlighted, collapsible JSON tree. | Structured data, API responses, extraction results. |
| Markdown | Rendered markdown with headings, lists, and links. | Reports, documentation, formatted analysis. |
| Tables | Interactive data grid with sorting and filtering. | Query results, aggregated metrics, comparison data. |
Exporting Results
You can export agent results from the results panel:
- Copy to clipboard -- Copy text, JSON, or markdown output.
- Download as file -- Export results as
.json,.md,.csv, or.txt. - Send to Data View -- Open table results in a Data View for further exploration.
- Share -- Share the result with colleagues via a link. Recipients see the result subject to their own access policies.
Run History
Every agent run is recorded in the run history. Access it from the History tab in the agent detail view or from AI > Agent Runs for a global view.
The run history shows:
| Field | Description |
|---|---|
| Run ID | Unique identifier for the run. |
| Status | running, completed, failed, cancelled. |
| Started at | Timestamp when the run began. |
| Duration | Total execution time. |
| Triggered by | The user, schedule, or event that initiated the run. |
| Input parameters | The values provided at execution time. |
| Output | The result, accessible in its original format. |
When a run fails, the history entry includes the error message, the step that failed, and the full execution log. Use this information to diagnose and fix issues in the agent specification.
Permissions
Access to the Agent Catalog is governed by your workspace role and any agent-level sharing settings:
- View -- All workspace members can browse the catalog and view agent details.
- Run -- Users with the Agent Operator role (or higher) can execute agents.
- Edit -- Only the agent author and users with the Agent Builder role can modify agent specifications.
- Delete -- Only workspace administrators and the agent author can delete agents.
Next Steps
- Agent Builder -- Create and configure custom agents.
- Workflow Builder -- Orchestrate agents in multi-step workflows.
- Multi-Agent Coordination -- Coordinate multiple agents with event-driven patterns.