Skip to content
Karyam

Runs

Runs represent the execution history of AI systems in Karyam.

Every time an Agent responds, an AI Flow executes, or an API request triggers work, a Run is created.

Runs provide the operational visibility required to safely operate AI systems in production.

Why Runs Exist

Section titled “Why Runs Exist”

AI systems are fundamentally different from traditional software.

Every execution may involve:

  • Multiple reasoning steps
  • Tool invocations
  • MCP calls
  • Human approvals
  • Retrieval operations
  • Workflow branching
  • External systems

Without visibility, AI systems become black boxes.

Runs make every execution observable and traceable.

What Creates a Run?

Section titled “What Creates a Run?”

Runs are automatically created whenever:

  • An Agent receives a request
  • An AI Flow is executed
  • A Service Catalog action is triggered
  • An API request starts execution
  • A scheduled workflow runs

Each Run represents a single execution instance.

What Information Does a Run Contain?

Section titled “What Information Does a Run Contain?”

A Run captures the complete execution lifecycle.

Typical information includes:

  • Job ID
  • Execution status
  • Start time
  • Duration
  • Model used
  • Token usage
  • Cost information
  • Execution logs
  • Output results

Run Lifecycle

Section titled “Run Lifecycle”

A typical run progresses through several stages.

Submitted
    ↓
Running
    ↓
Completed

Possible outcomes include:

  • Success
  • Failure
  • Cancellation
  • Waiting Approval
  • Paused

Run Statuses

Section titled “Run Statuses”

Karyam tracks the state of each execution.

Common statuses include:

Status Description
RUNNING Execution is currently in progress
SUCCESS Execution completed successfully
FAILED Execution encountered an error
WAITING_APPROVAL Waiting for human approval
PAUSED Execution temporarily paused
CANCELLED Execution was cancelled

Run Information

Section titled “Run Information”

Each run contains operational metadata.

Examples include:

Job Information

Section titled “Job Information”
Job ID
Flow
Agent
Execution Duration
Start Time
Status

Model Information

Section titled “Model Information”
Model Used
Temperature
Maximum Steps
Caching Status

Usage Information

Section titled “Usage Information”
Input Tokens
Output Tokens
Total Tokens
Input Cost
Output Cost
System Cost

This allows organizations to monitor both usage and spending.

Logs

Section titled “Logs”

Every run generates execution logs.

Logs provide visibility into:

  • Agent reasoning
  • Tool invocations
  • MCP calls
  • Workflow steps
  • Approvals
  • Errors

This makes debugging significantly easier.

Sub Runs

Section titled “Sub Runs”

Complex AI systems often trigger additional executions.

Examples include:

Main Flow
├── Research Agent
├── Approval Workflow
└── Ticket Creation Agent

These child executions appear as Sub Runs.

This allows teams to understand how work was delegated throughout the system.

API Executions

Section titled “API Executions”

Runs are created regardless of how execution was triggered.

Examples include:

  • Web Interface
  • Service Catalog
  • API Calls
  • Schedules
  • Listeners

Everything ultimately becomes a Run.

Relationship to Other Concepts

Section titled “Relationship to Other Concepts”
Agent
    ↓
AI Flow
    ↓
Execution
    ↓
Run
    ↓
Logs

Runs sit at the center of operational visibility within Karyam.

Runs and Observability

Section titled “Runs and Observability”

Runs work together with:

  • Agent Logs
  • Workflow Logs
  • Token Usage
  • Cost Analytics
  • Audit Logs

Together they provide complete operational observability.

The Karyam Philosophy

Section titled “The Karyam Philosophy”

Production AI requires more than execution.

It requires visibility.

Runs provide the foundation for debugging, governance, auditing, and operational excellence.