AI Agents

The Agents system lets you create, configure, and run AI-powered automation processes. Describe what you want to automate in plain English using the AI Agent Builder, or dive into full control with the visual drag-and-drop workflow editor. Browse pre-built templates, deploy agent instances with custom configurations, and monitor execution results — all from a unified interface.

Two Ways to Build Agents

The platform offers two complementary approaches to agent creation:

Approach	Best For	Where to Find It
AI Agent Builder	Quick creation using natural language — no technical knowledge required	Dashboard → Agents tab → Agent Builder subtab, or via the Agent Command pill
Visual Workflow Builder	Full control over node selection, connections, and configuration	Sidebar → Agents → Workflows tab

Start with the AI Agent Builder when you know what you want to automate but not how. Switch to the visual builder when you need fine-grained control over individual nodes and data flow.

AI Agent Builder

The AI Agent Builder is a conversational interface that turns natural language descriptions into fully configured agents. It is the fastest way to go from idea to running automation.

How It Works

1. Describe your goal — type what you want to automate in the message box. For example: "Monitor our knowledge base for new compliance documents and send weekly summary reports."
2. AI understands your intent — the builder classifies your request and identifies the automation pattern.
3. Review suggestions — the builder either matches your request to an existing template or generates a custom workflow from scratch.
4. Preview the workflow — a visual graph of the proposed nodes and connections appears alongside a configuration summary.
5. Create or refine — click Create Agent to deploy immediately, or continue the conversation to adjust requirements.

Conversational Refinement

The builder supports multi-turn conversation. After the initial suggestion, you can:

• Add constraints ("only process PDFs uploaded in the last 7 days")
• Change the output format ("send results as a Slack message instead of email")
• Request alternatives ("show me a simpler version with fewer steps")
• Ask questions ("what does the Data Collector node do?")

Each follow-up refines the workflow preview in real time.

Smart Defaults

The builder applies intelligent defaults based on your tenant's usage patterns and configuration history. If your team frequently uses specific knowledge base categories, LLM models, or output formats, the builder pre-fills those settings so you spend less time configuring and more time automating.

Model Selection

Toggle between two AI modes using the model selector above the conversation:

• Smart — higher quality understanding and generation, uses more tokens
• Fast — quicker responses at lower cost, suitable for straightforward requests

After Creation

Once you create an agent from the builder, you can:

• Run it immediately from the Active Agents view
• Open it in the visual builder to fine-tune individual nodes and connections
• Schedule it for recurring execution on the Agents management page
• Promote the workflow to a template so others on your team can reuse it

Agent Command Pill

The Agent Command pill is a floating button at the bottom-left of your screen, available on every page. It provides quick access to agent features without navigating away from your current task.

What You Can Do

• Quick agent creation — type a short description and press Enter to jump directly into the AI Agent Builder with your prompt pre-loaded
• See running agents — a badge on the pill shows how many agents are currently executing
• Navigate quickly — use the quick-action links to jump to the Agent Builder, Active Agents, or Templates view

Tip: The Agent Command pill appears alongside the Intelligence pill. Both are visible when the intelligence layer is enabled on your subscription.

Getting Started

Open Agents from the sidebar for full management, or use the Agents tab on the main dashboard for the streamlined AI-first experience. The agents system requires the agents_basic capability on your subscription and the AGENTS_READ permission on your role. Additional capabilities unlock more features:

Capability	What It Unlocks
agents_basic	Browse templates, view agent configurations, use the AI Agent Builder
agents_enterprise	Create and manage agent instances, monitor executions
agents_advanced	Build custom workflows with the visual editor

You also need an active infrastructure stack with vector database, storage, and worker resources provisioned for your tenant.

Agents Management Page

The dedicated Agents page (accessible from the sidebar) is organized into tabbed sections for full management. The tabs you see depend on your capabilities:

Tab	Description	Required Capability
Templates	Browse, search, and manage agent templates	agents_basic
Instances	Deploy and manage running agent instances	agents_enterprise
Monitor	Track execution status, view logs, and review results	agents_enterprise
Workflows	Build custom workflows with the visual node editor	agents_advanced

A status bar at the top of the page shows the current infrastructure stack and the number of active plugin integrations and available node types.

Templates

Templates are the blueprints for AI agents. Each template defines a workflow of connected processing nodes, default parameters, and the tools and plugins the agent needs.

Browsing Templates

The Templates tab displays templates in three sub-tabs:

• All Templates — every template available to your tenant, including system-provided and custom templates
• System Templates — pre-built templates provided by the platform, covering common automation patterns
• My Templates — templates you created or that were promoted from your custom workflows

Use the search bar and filters to narrow results by name, category, or complexity level (Simple, Medium, Complex, Enterprise).

Template Categories

Templates are organized by business function. Common categories include data analysis, document processing, research automation, compliance, customer support, and reporting. Select a category filter to focus on templates relevant to your workflow.

Template Details

Click any template card to view its details:

• Description — what the template does and its intended use case
• Category and complexity — classification and expected resource intensity
• Required plugins — which platform plugins must be active for this template
• Required capabilities — the minimum capability level needed to use it
• Default parameters — pre-configured settings that you can override when creating an instance
• Estimated duration — approximate execution time for a typical run
• Compatible stacks — which infrastructure stacks support this template

Creating a Template

Tenant administrators and superusers can create new templates from scratch or promote existing workflows to templates. To create a template directly:

1. Click Create Template in the Templates tab
2. Enter a name, description, and category
3. Set the complexity level and visibility (public or tenant-specific)
4. Define the workflow definition, default parameters, and required plugins
5. Save the template

Tip: The easiest way to create a template is to build and test a workflow first, then promote it to a template from the Workflows tab.

Managing Templates

Depending on your permissions, you can perform the following actions on templates:

• Edit — update the template name, description, category, complexity, or visibility
• Regenerate schema — rebuild the configuration schema from the workflow definition
• Soft delete — remove a template from active view (can be restored later)
• Restore — bring back a previously deleted template
• View webhooks — manage inbound webhook triggers attached to this template

Superusers can manage system-level templates. Tenant admins can manage templates scoped to their tenant and stack.

Agent Instances

An agent instance is a deployed copy of a template, configured for your specific use case. Instances hold the runtime configuration, scheduling settings, and execution history.

Creating an Instance

1. Open a template and click Create Instance (or click the create icon on the template card)
2. Enter a name and optional description
3. Configure the instance settings using the schema-driven configuration editor — each template defines its own set of configurable parameters
4. Save the instance

The instance appears in the Instances tab, ready to run.

Instance Configuration

The configuration editor adapts to each template's schema. Common configuration areas include:

• Execution preferences — timeout duration, retry behavior, concurrency limits
• Node overrides — customize settings for individual processing nodes in the workflow
• Data categories — restrict which knowledge base categories the agent can access
• Allowed users — limit who can execute this instance

You can modify an instance's configuration at any time. Changes take effect on the next execution.

Running an Instance

There are several ways to execute an agent instance:

• Run — execute immediately with current configuration from the Instances tab
• Execution Client — open a full-screen interactive execution view with message input, streaming output, and real-time progress updates via WebSocket
• Schedule — set up automatic recurring execution using a cron-based schedule
• Webhook — trigger execution from external systems via inbound webhooks
• API — trigger execution programmatically through the REST API

Scheduling

To set up a recurring schedule for an instance:

1. Select the instance and click Schedule
2. Use the schedule builder to define a cron expression or pick a preset (hourly, daily, weekly)
3. Optionally configure notification settings for completed or failed runs
4. Save the schedule

Once scheduled, the instance shows a schedule indicator. You can pause, resume, or remove the schedule at any time. View past scheduled runs in the execution history.

Managing Instances

Available actions on instances:

• Configure — open the configuration editor to modify settings
• View workflow — inspect the underlying workflow diagram in a read-only viewer
• Stop — cancel a currently running execution
• Soft delete / Restore — archive or recover instances

Workflows

The Workflows tab provides a visual builder for creating custom AI workflows from scratch. Workflows define how data flows through processing nodes to achieve automated outcomes.

Workflow List

The Workflows tab shows all workflows in your tenant with:

• Name and description
• Status — Draft, Active, or Inactive (click the status chip to change it)
• Category — business function classification
• Plugin integrations — which connected plugins the workflow uses
• Validation warnings — any structural issues detected by the validator

Creating a Workflow

1. Click Create Workflow
2. Enter a name and description
3. The Workflow Builder opens in a full-screen dialog

Workflow Builder

The visual builder is a canvas-based editor where you design agent workflows by connecting processing nodes:

Node Palette

Open the node drawer to browse available node types. Nodes are grouped by function:

• Core nodes — LLM Processor, Web Search, Conditional Branch, Database Query, Data Collector, Analyzer, Custom Function, and parallel processing nodes (Fan Out/Fan In, Multi-Source Collector, Parallel Orchestrator)
• Plugin nodes — nodes that integrate with other platform plugins (Knowledge, Files, Reports, Notifications, Scheduler, Integrations, Chat, Organizations, Audit, API Management)
• Transformer nodes — data format converters (Text to JSON, JSON to Text, and others)

Drag a node from the palette onto the canvas to add it.

Connecting Nodes

Draw edges between node output ports and input ports to define data flow. The builder validates connections in real time:

• Port type compatibility is checked automatically
• Invalid connections display a warning
• Use Validate Workflow to run a full structural check across all nodes and edges

Node Configuration

Click a node on the canvas to open its configuration form. Each node type has a schema-driven settings panel covering:

• Input parameters and output mappings
• Processing-specific options (e.g., LLM model selection, search parameters, query templates)
• Connection bindings for integration nodes

Node Testing

Right-click a node and select Test Node to execute it in isolation with sample data. The test dialog lets you provide mock inputs and review the node's output without running the entire workflow.

Saving and Executing

• Save — persist the current workflow definition
• Execute — run the workflow immediately. You can choose the execution mode and provide runtime input data
• Execute Ad-Hoc — run an unsaved workflow definition directly for quick testing

Important: Unsaved changes trigger a navigation guard. Save your workflow before leaving the builder to avoid losing work.

Promoting a Workflow to a Template

When a workflow is tested and stable, you can promote it to a reusable agent template:

1. Ensure the workflow status is Active
2. Click Promote to Template from the workflow actions menu
3. The platform creates a new template from the workflow definition, preserving all nodes, edges, and configuration

The promoted template appears in the Templates tab and can be used to create agent instances.

Workflow Validation

The built-in validator checks your workflow for:

• Disconnected nodes or missing connections
• Port type mismatches between connected nodes
• Missing required configuration on nodes
• Circular dependencies

Run validation before executing or promoting a workflow to catch issues early.

Execution Monitor

The Monitor tab provides a real-time dashboard of all agent executions across your tenant.

Execution List

View all executions in a sortable, filterable grid:

Column	Description
Agent	The instance or workflow that ran
Status	Pending, Running, Completed, Failed, or Cancelled
Type	How the execution was triggered (manual, scheduled, webhook, API)
Started	When the execution began
Duration	Total execution time
Tokens	Number of AI tokens consumed

Use the status filter dropdown to focus on specific execution states. Enable Auto-refresh to poll for updates on running executions automatically.

Execution Details

Click any execution to open its detail dialog:

• Summary — status, timing, trigger type, and resource usage
• Node results — output from each processing node in the workflow
• Logs — step-by-step execution trace with timestamps
• Knowledge sources — which knowledge base documents were consulted
• LLM debug — raw model interactions for troubleshooting AI responses
• Error details — full error messages and stack traces for failed runs

Execution Statistics

Access aggregate statistics to understand execution patterns:

• Total runs, success rate, and average duration
• Token consumption over time
• Most-used templates and workflows

Webhook Triggers

Inbound webhook triggers let external systems start agent executions automatically. Each trigger maps a webhook connection from the Integrations plugin to an agent template.

Setting Up a Webhook Trigger

1. Open a template and click Webhooks (or manage from the template actions menu)
2. Click Create Trigger
3. Select a webhook connection from your integrations
4. Configure event type filters and payload mapping
5. Set the execution mode (synchronous or asynchronous) and concurrency limits
6. Activate the trigger

When the webhook connection receives a matching event, the platform automatically creates and runs an agent execution using the linked template.

Managing Triggers

View trigger statistics including total events received, successful triggers, and error counts. Deactivate triggers temporarily without deleting them.

Best Practices

• Start with the AI Agent Builder — describe your goal in natural language first. The builder suggests templates or generates workflows automatically, saving you the effort of manual node selection
• Start with system templates — explore the pre-built templates before creating custom workflows. They cover common patterns and are optimized for performance
• Test workflows incrementally — use node testing to validate individual nodes before connecting them into complex graphs
• Use categories for access control — assign data categories to instances to control which knowledge base content agents can access
• Monitor token usage — check execution statistics regularly to understand AI resource consumption. The AI Agent Builder's token usage is also tracked and counted towards your quota
• Set appropriate timeouts — configure timeout values on instances to prevent runaway executions
• Schedule during off-peak hours — for resource-intensive agents, schedule runs during low-usage periods
• Promote tested workflows — only promote a workflow to a template after thorough testing to ensure reliability
• Use the Agent Command pill — access agent features from any page without interrupting your workflow

Troubleshooting

Agent page shows a capability warning

Problem: You see a message that agents_basic capability is required. Solution: Contact your administrator to enable the agents capability on your subscription. The agents plugin requires at least agents_basic to access the Templates tab.

Cannot create instances or see the Monitor tab

Problem: The Instances and Monitor tabs are not visible. Solution: These tabs require the agents_enterprise capability. Contact your administrator to upgrade your subscription.

Workflow builder does not show plugin nodes

Problem: Only core nodes appear in the node palette. Solution: Plugin-integrated nodes require the corresponding plugin to be active on your tenant. Verify that the relevant plugins (Knowledge, Files, Reports, etc.) are enabled in your subscription.

Execution fails with a timeout error

Problem: An execution stops with a timeout status. Solution: Open the instance configuration and increase the timeout value. For complex workflows with many nodes, consider increasing from the default. Also check that external services (APIs, databases) are responding within expected times.

Webhook trigger does not fire

Problem: External events arrive but no execution starts. Solution: Verify the trigger is active, the event type filter matches the incoming event, and the payload mapping is correct. Check the webhook event log for received events and any processing errors.

Scheduled execution did not run

Problem: A scheduled agent instance did not execute at the expected time. Solution: Verify the schedule is active (not paused). Check that the instance status is not stopped or deleted. Review the execution history for any failed attempts. Ensure the worker resources in your infrastructure stack are running.

Related Topics

• Knowledge Management — User Guide
• AI Chat — User Guide
• Dashboard Navigation Guide
• Platform Features Overview

---

⏱️ Read time: 18 minutes | 📊 Difficulty: intermediate | 🔄 Last updated: 2026-04-08