Agents

April 4, 2026 · View on GitHub

What Is This?

Agents are the platform's AI executors. They are the ones who actually "do the work" — reading code, writing code, running tests, submitting PRs. You don't need to manually instruct them at every step; just create a ticket, and the agent will execute automatically according to the Workflow.

Key Concepts

ConceptDescription
Agent DefinitionA project-level configuration binding an AI Provider (e.g. Claude Code) with project parameters
Agent RunA complete execution instance of an agent working on a ticket
Agent OutputLogs and results produced during execution
Agent StepHuman-readable action stage descriptions

Supported AI Providers

ProviderSource
Claude CodeAnthropic
CodexOpenAI
Gemini CLIGoogle

Common Operations

Registering an Agent

  1. Go to the Agents page
  2. Select from available Providers
  3. Name the agent and confirm creation

Monitoring Agents

  • The sidebar shows a badge with the count of active agents
  • Click through to view all agent statuses (Active / Paused / Retired)
  • Click a specific agent to see its current tickets and execution history

Agent Lifecycle Management

ActionDescription
PauseTemporarily stop the agent from picking up new tickets
ResumeResume from paused state
RetirePermanently deactivate an agent

Real-time Execution Viewing

Agent execution supports real-time streaming output (SSE). In the Agent Run detail page you can see:

  • Step-by-step operation descriptions
  • Live log output
  • Execution status changes

Tips

  • A single project can have multiple agents, each bound to different workflows (e.g. one for coding, one for testing)
  • If an agent behaves abnormally, first check the Machine connection status and Workflow Harness configuration
  • Agent execution history helps you understand AI decision-making and optimize Harness instructions