lantern play

lantern play is AgentLantern's live runtime viewer. It runs a supported agent project and opens a pixel-art UI where you can watch agents work, think, delegate, call tools, and produce a final report.

Use this section when you want to:

• launch a supported agent project from the UI,
• understand the Play interface,
• verify how agents are placed in the world,
• inspect tool calls through the Tool Hub,
• save and replay a run,
• debug why a run or UI does not update.

lantern play executes the target project. The project can call LLM providers, APIs, files, tools, and network services configured by that project.

Pages

Page	Use it for
Quick Start	Launch your first run and understand the ports
Interface	Learn the roster, city, activity panels, bubbles, and report view
Agent Layouts	Understand how 1-10+ agents are positioned
Tool Hub	Understand how tools are visualized
Replay	Save and replay runs
Command Reference	See CLI options and runtime capture details
Troubleshooting	Fix START, stale UI, ports, blocked runs, and stopping

Mental Model

lantern play has three moving parts:

Part	Role
Play server	Serves the UI and WebSocket event stream
Child process	Runs the target project from its own root
Browser UI	Displays agents, timeline, logs, tools, thoughts, and report output

The UI does not auto-start by default. You start the run deliberately from the browser with START.

Replay From The Same UI

lantern replay uses the same Play environment, but it replays a saved event stream instead of running the agent project again.

# First run and save a replay
lantern play /path/to/project --name demo-run

# Later, replay the saved run
lantern replay demo-run

# Or replay the newest saved run
lantern replay last

From inside the project root, this is enough:

lantern play
lantern replay demo-run
lantern replay last

When lantern play starts without --name, the UI asks for a run name before START and saves the replay with that name. lantern replay then opens the same pixel-art world with replay controls: pause, seek, speed, restart, and stop.

Use replay when you want to demo or debug a run without spending tokens or calling external tools again.

Current Runtime Support

Framework	Status
CrewAI	Live runtime instrumentation
Google ADK	Runner events and function calls
LangGraph	Graph invocation lifecycle
AutoGen	Team stream, messages, tools and usage
Smolagents	Agent task lifecycle
OpenAI Agents SDK	Native tracing processor for agents, tools, handoffs and usage
Additional Python ecosystems	Agent invocation lifecycle, stdout and static roster
Mastra / Vercel AI SDK / other JS/TS	npm execution, process lifecycle, stdout and static roster
Semantic Kernel .NET/Java	dotnet, Maven or Gradle execution, stdout and static roster
Other frameworks	Process lifecycle, stdout and static agent roster

Runtime capture is best effort and depends on the framework version and execution API. Use --command when the project has a custom launcher.