Understanding Agents

In Supervity, agents are AI-powered operators responsible for planning, executing, and supervising work on behalf of humans.

Agents do not simply respond to prompts.
They act as decision-making systems that operate within clearly defined permissions, policies, and human controls.

They are the primary abstraction through which all work in Supervity is delegated.

What Agents Are (and Aren’t)

Agents are:

• Autonomous but governed
• Capable of multi-step reasoning
• Responsible for planning execution
• Able to use tools and integrations
• Designed to pause for human input
• Fully observable and auditable

Agents are not:

• Simple chatbots
• Prompt-only assistants
• Static workflows
• Hardcoded scripts
• Black-box automation

Agents reason first, execute second.

The Agent Execution Model

Agents operate through a consistent lifecycle that applies to all roles and use cases.

1. Intent Understanding

A human defines what outcome is desired.

Example:

“Every weekday, review new support tickets and escalate urgent ones.”

The agent:

• interprets intent
• identifies required systems
• detects missing information
• asks clarifying questions when needed

No execution begins at this stage.

2. Planning & Reasoning

Once intent is clear, the agent constructs a plan that defines:

• triggers (schedule, event, API)
• steps required to complete the job
• conditional logic and branching
• approval and review checkpoints
• retry and failure behavior

This plan is:

• visible
• reviewable
• modifiable

Humans always approve the plan before execution.

3. Jobs & Workflows

When execution begins, the agent creates a job.

• A job represents a single execution of work.
• The job runs through a deterministic workflow.
• The workflow defines execution order, dependencies, and logic.

Agents plan jobs.
Workflows execute jobs.

→ Learn more: Workflows Explained

4. Tool & Integration Usage

During execution, agents may use tools to act on systems.

Tools include:

• built-in integrations (email, CRM, files, chat)
• registered external APIs
• internal services
• custom tools

Tool usage is:

• explicitly permissioned
• scoped per workspace
• logged and auditable

Agents decide when to use tools.
The platform governs how tools are accessed.

→ Learn more: Integrations Overview

5. Human-in-Command Oversight

For sensitive or high-impact steps, agents pause execution.

Examples:

• approval required
• manual input needed
• exception handling
• decision escalation

When a review step is reached:

• the job pauses safely
• full context is preserved
• execution resumes only after a human decision

→ Learn more: Human Review & Approvals

6. Execution, Monitoring & Adaptation

Once approved, agents:

• execute steps sequentially or in parallel
• retry safely on failure
• emit logs, metrics, and events
• surface progress and outputs in real time

Jobs can be triggered via:

• manual runs
• schedules
• webhooks
• APIs

Agents adapt execution based on outcomes and human input.

What Agents Can Do

Orchestrate Workflows

• build multi-step execution plans
• coordinate across multiple systems
• apply conditional logic
• handle failures gracefully

Process Data & Content

• extract structured data
• transform formats
• summarize or generate content
• route outputs downstream

Support Decision-Making

• evaluate conditions
• classify or score inputs
• recommend actions
• defer final authority to humans

Operate via APIs

• act as backend execution primitives
• trigger jobs programmatically
• support headless operation
• embed into external systems

→ See details: API Reference

Agent Ownership & Access Control

Agents exist within a workspace and respect role-based access.

• Admins
  Configure permissions, tools, and policies

• Editors
  Create and modify agents and workflows

• Runners
  Execute jobs

• Viewers
  Observe runs and logs

Every action is logged and auditable.

Saved Agents & Reuse

Agents are persistent system entities.

They can be:

• saved and reused
• shared across teams
• scheduled to run automatically
• triggered externally
• versioned and rolled back

Agents evolve over time.
They are not one-off automations.

Security & Governance

Agents operate under strict controls:

• least-privilege permissions
• encrypted credentials
• policy-based restrictions
• approval gates
• full audit trails

An agent cannot act outside what a human has explicitly allowed.

Who Uses Agents?

Business Users

• delegate operational work
• stay in control through approvals

Operations Teams

• standardize execution
• enforce governance and reliability

Developers & Platform Teams

• use agents as execution units
• integrate via APIs
• extend capabilities with tools

Where to Go Next

• Understand execution → Workflows Explained
• Add governance → Human Review & Approvals
• Integrate systems → Integrations Overview
• Trigger programmatically → API Reference

Agents are the core system primitive in Supervity.
Everything else exists to support how they plan, execute, and collaborate with humans.