Framework Coverage

AgentLantern 0.4.0 uses dedicated analyzers where an SDK exposes recognizable architecture, then falls back to conventional agent patterns for custom Python and JavaScript/TypeScript projects.

lantern detect .    # explain which adapter matched and why
lantern inspect .   # print the normalized project model
lantern docs .      # generate architecture documentation
lantern lint .      # run deterministic checks
lantern play .      # launch the project in the Play UI

Dedicated adapters

Ecosystem	Framework	Static model	Play capture
Python	CrewAI	Agents, tasks, tools, delegation, configuration	Deep lifecycle, tools, delegation, output
Python	LangGraph	Graphs, nodes, routes, tools, invocations	Graph invocation lifecycle
Python	AutoGen	Agents, teams, tools, streams	Messages, tools, usage, team lifecycle
Python	Smolagents	Agents, managed agents, tools	Agent task lifecycle
Python	Google ADK	Root agents, sub-agents, tools, workflows	Runner events and function calls
Python	OpenAI Agents SDK	Agents, handoffs, tools, guardrails, runners	Native tracing processor
Python	PydanticAI	Agents, models, prompts, decorated tools, runs	Agent invocation lifecycle
Python	LlamaIndex	Function/ReAct agents, workflows, tools	Agent/workflow lifecycle
Python	Semantic Kernel	Agent types, plugins, orchestrations	Invocation lifecycle
Python	Agno	Agents, teams, tools, instructions	Run lifecycle
Python	LangChain Agents	Agent factories, tools, executors, invocations	Runnable lifecycle
Python	Haystack Agents	Agents, tools, pipelines, runs	Agent lifecycle
Python / TypeScript	Strands Agents	Agents, tools, swarms, graphs, workflows	Python hooks; Node process and logs
Python	CAMEL-AI	Chat agents, workforces, role-playing	Agent step lifecycle
Python	MetaGPT	Role subclasses, teams, runs	Role lifecycle
Python / TypeScript	BeeAI	ReAct/tool-calling agents and workflows	Python hooks; Node process and logs
Python	OpenAI Swarm (legacy)	Agents, functions, transfers, client runs	Swarm lifecycle
TypeScript	Mastra	Agents, tools, workflows, npm scripts	Process, roster, logs, result
TypeScript	Vercel AI SDK Agents	ToolLoopAgent, models, tools, generate/stream	Process, roster, logs, result
.NET / Java	Semantic Kernel	Agent initializers, instructions, invocations	Process, roster, logs, result

Generic adapters

The generic analyzers are checked after every dedicated adapter.

Python

Recognized patterns include:

• Agent(...) and classes ending in Agent;
• subclasses of Agent, BaseAgent, or AbstractAgent;
• create_agent(...), build_agent(...), make_agent(...), and similar factories;
• common tools, handoffs, managed_agents, teams, workflows, and run methods.

JavaScript and TypeScript

Recognized patterns include:

• new Agent({...}) and new CustomAgent({...});
• createAgent({...}), create_agent({...}), buildAgent({...}), and makeAgent({...});
• tool objects, teams, workflows, swarms, and common execution methods.

Runtime fidelity

Static support and live runtime capture are separate capabilities. Python frameworks with public hooks expose richer Play events. Node, .NET, Java, and unknown frameworks retain the static roster, command lifecycle, stdout, errors, replay, and final output.

Use an explicit launcher when project conventions are unusual:

lantern play . --command "your command here"

Extend AgentLantern

An internal framework with unusual APIs can publish a BaseAnalyzer through the agentlantern.analyzers Python entry-point group. Dedicated analyzers take precedence over the generic fallbacks.