Supported Coding Agents
Operator orchestrates AI coding agents to work on tickets from your kanban board. The following agents are currently supported:
Available Agents
| Agent | Status | Notes |
|---|---|---|
| Claude | Recommended | Full feature support |
| Codex | Supported | OpenAI’s coding model |
| Gemini CLI | Experimental | Google’s AI assistant |
Agent Capabilities
Each agent can:
- Read and understand ticket requirements
- Browse project codebases
- Write and modify code
- Run tests and builds
- Create pull requests
Choosing an Agent
Claude is recommended for most users due to its strong code understanding and generation capabilities. See individual agent pages for setup instructions and specific features.
Agent Lifecycle
Operator tracks every agent it launches through these states:
Created -> Running -> Completed
|
v
Awaiting Input
| State | Description |
|---|---|
| Created | Agent initialized, not yet started |
| Running | Actively working on a ticket |
| Awaiting Input | Needs a human response |
| Completed | Work finished successfully |
| Failed | An error occurred |
Autonomous and Paired Modes
Every issue type declares a mode, and that decides how much of your attention
its tickets need:
- Autonomous — launch and monitor. Minimal intervention, and several can run in parallel across different projects.
- Paired — active human participation, with back-and-forth discussion. One at a time, because they compete for the same operator: you.
Mode is a property of the issue type, not of the agent, so a collection decides which of its work types are hands-off. See Workflows.
Sessions
Agent sessions persist under .operator/:
.operator/
├── state.json
├── sessions/
│ ├── agent-123.json
│ └── agent-456.json
└── history.json
Session files record ticket information, start and end times, status history, and output logs. Operator can also detect completion from files an agent produces — see Artifact Detection.
Best Practices
- Monitor paired agents — stay engaged with paired work
- Review autonomous work — check completed tickets
- Handle failures promptly — address failed agents quickly
- Balance load — don’t overload with too many agents