Build / Agents
Let AI run the work between your tools.
Agents turn business intent into repeatable action. They respond to events, understand context, use tools, make decisions, and move work forward across your applications.
From trigger to outcome.
An agent observes a trigger, gathers the right context, follows its instructions, makes a decision, and takes action. Every run can be inspected so automation stays understandable and accountable.
Two ways to build
Choose flexible reasoning or visible control.
Start with the type of agent that matches the work. Use prompts for instruction-led tasks and flows for processes that need an explicit sequence.
Prompt agents
Give the agent a clear instruction and the context it needs. Prompt agents are useful for summarizing requests, qualifying leads, reviewing records, drafting updates, and deciding what should happen next.
Flow agents
Connect visible blocks into an inspectable workflow. Flow agents are useful when a process needs defined steps, conditions, branches, record operations, notifications, or external calls.
Start the work
Agents can respond to the moments that matter.
Configure an agent to run when work begins, changes, or needs to happen again.
Manual input
Run an agent on demand and pass it the context for the task.
Record events
Start when a record is created, updated, or deleted in a configured table.
Schedules
Run recurring work daily, weekly, monthly, or at a configured time.
Social comments
React to configured social comment events when the required channel connection is available.
Agent building blocks
Compose the work one decision at a time.
Use focused blocks to give an agent the context, decisions, and actions it needs to complete a process.
Read and change records
Read business data, add new records, update existing records, or delete records when the configured permissions allow it.
Check conditions
Branch the workflow based on a value, decision, or result so different situations can follow different paths.
Process context
Use prompt blocks to ask the model to interpret the context and return a useful result for the next step.
Take action
Schedule work, send email notifications, or call a configured external POST endpoint.
Example workflow
When a new lead arrives, move the next step forward.
“When a new lead enters CRM, qualify it and notify the owner.”
This request can become a connected workflow that reads the lead, evaluates its intent and fit, and gives the owner a concise handoff summary.
1. Lead created
The record event starts the agent for the configured CRM table.
2. Read details
The agent fetches the contact, company, and other relevant business data.
3. Qualify
A prompt or decision block evaluates intent, fit, and the recommended next action.
4. Notify owner
The agent sends a concise handoff summary so the owner can act on the result.
Execution history
Automation should never become a black box.
Review each run to understand what happened, which blocks were involved, and where an outcome succeeded or failed.
Run context
See the trigger type, start time, execution status, and execution ID for each run.
Block activity
Inspect the blocks that started, completed, or failed during the execution.
Tool activity
Review the tools and actions used while the agent processed its task.
Failure detail
Use execution details to understand a failure and refine the agent definition.
Agent lifecycle
Create, configure, test, and improve.
An agent becomes useful through an iterative operating loop.
1. Create
Choose a name, description, and agent type: prompt or flow.
2. Configure
Define the instructions, trigger, blocks, connections, and permissions.
3. Test
Run the agent manually, inspect the result, and review its execution history.
4. Operate
Activate, deactivate, update, and improve the agent as the business process changes.
Controls and boundaries
Give agents the right tools—and the right limits.
Agents act within their configured definition, available tools, workspace permissions, connected credentials, and active status. Custom agents can be updated or deleted. Core system agents are protected and cannot be modified or deleted.