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
| Concept | Description |
|---|---|
| Agent Definition | A project-level configuration binding an AI Provider (e.g. Claude Code) with project parameters |
| Agent Run | A complete execution instance of an agent working on a ticket |
| Agent Output | Logs and results produced during execution |
| Agent Step | Human-readable action stage descriptions |
Supported AI Providers
| Provider | Source |
|---|---|
| Claude Code | Anthropic |
| Codex | OpenAI |
| Gemini CLI |
Common Operations
Registering an Agent
- Go to the Agents page
- Select from available Providers
- 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
| Action | Description |
|---|---|
| Pause | Temporarily stop the agent from picking up new tickets |
| Resume | Resume from paused state |
| Retire | Permanently 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