Play Command Reference
lantern play
bash
lantern play [project] [--ws-port PORT] [--http-port PORT] [--name NAME]| Option | Default | Description |
|---|---|---|
project | current directory | Agent project to run |
--ws-port | 7890 | WebSocket port for live events |
--http-port | 7891 | HTTP port for the Play UI |
--name | empty | Save the run as .lantern_replays/<name>.jsonl and auto-start. If omitted, the UI asks for the run name before START |
lantern replay
bash
lantern replay NAME [--speed FLOAT] [--ws-port PORT] [--http-port PORT]| Option | Default | Description |
|---|---|---|
NAME | required | Replay name, direct .jsonl path, or last for the newest saved replay |
--speed | 1.0 | Playback speed multiplier |
--ws-port | 7890 | WebSocket port for replay events |
--http-port | 7891 | HTTP port for the replay UI |
Runtime Events
AgentLantern instruments the child process with a framework-specific adapter and captures:
| Event type | Description |
|---|---|
| Crew lifecycle | crew start, crew done, process messages |
| Tasks | task start, task done, task error |
| Thoughts | agent step or callback output when available |
| Delegation | coworker questions and delegated work |
| Tools | framework tool/function calls when exposed by the runtime |
| stdout | raw terminal output from the child process |
Adapters are included for CrewAI, LangGraph, Google ADK, AutoGen, Smolagents, and the OpenAI Agents SDK. Unknown Python frameworks retain process lifecycle, stdout, and the statically extracted roster.
For a standard CrewAI scaffold, the target command is usually:
bash
uv run run_crewAgentLantern detects and runs the appropriate command from the project root.