Usage Guide

Commands

lantern lint [project] [--strict] [--format text|json|sarif]

Analyzes an agent project for design issues without using an LLM.

# From the project root
cd my-agent-project
lantern lint

# From anywhere
lantern lint /path/to/my-agent-project

# Exit with code 1 on any finding (CI-friendly)
lantern lint --strict

# Machine-readable output for CI pipelines and integrations
lantern lint --json
lantern lint --format sarif > agentlantern.sarif

See the Linter reference for all rules and codes.

lantern inspect [project]

Outputs the full project model as JSON for integrations and debugging.

lantern inspect /path/to/my-agent-project

lantern play [project] [--ws-port PORT] [--http-port PORT] [--name NAME] [--command CMD]

Runs a supported agent project and opens the live animated Play UI.

# From the project root
cd my-agent-project
lantern play

# From anywhere
lantern play /path/to/my-agent-project

# Custom ports
lantern play /path/to/my-agent-project --ws-port 8790 --http-port 8791

# Save the run for replay and auto-start it
lantern play /path/to/my-agent-project --name demo-run

# Override entrypoint auto-detection without invoking a shell
lantern play . --command "uv run my-agent-command"

The browser UI starts in a ready state. If --name is omitted, the UI asks for a run name before START. Use STOP to terminate the active run.

The Play UI includes dynamic 1-10 agent layouts, a central Tool Hub for tool calls, timeline/thoughts/tools/comms/log panels, clickable agent history, and final report.md display when available.

See the Play reference.

lantern replay NAME [--speed FLOAT]

Replays a saved lantern play --name run.

lantern replay demo-run
lantern replay last
lantern replay demo-run --speed 2.0
lantern replay /path/to/.lantern_replays/demo-run.jsonl

lantern compare BEFORE AFTER [--json]

Compares runtime, errors, agents, tools, token usage, and cost fields recorded in two replays.

lantern compare baseline candidate
lantern compare baseline candidate --json > replay-diff.json

lantern docs [project] [-o OUTPUT_DIR] [--dry-run] [--force] [--clean]

Generates documentation for an agent project.

# From the project root
cd my-agent-project
lantern docs

# From anywhere
lantern docs /path/to/my-agent-project

# Custom output directory
lantern docs /path/to/my-agent-project -o /path/to/output

# Safe preview and one-time adoption of output from AgentLantern < 0.2
lantern docs --dry-run
lantern docs --force --clean

lantern web [project]

Generates (unless --no-generate) and serves the documentation for an agent project over local HTTP.

# From the project root
cd my-agent-project
lantern web

# From anywhere
lantern web /path/to/my-agent-project

# Custom port
lantern web --port 9000

# Serve without regenerating
lantern web --no-generate

# Accessible on the local network
lantern web --host 0.0.0.0

By default, lantern web serves http://localhost:9000 over local HTTP, so the browser should not show a certificate warning.

Supported Frameworks

AgentLantern detects the framework automatically from pyproject.toml dependencies or project structure.

Framework	Detection signal	Analysis level
CrewAI	crewai dep or crew.py + @CrewBase	Full — agents, tasks, diagrams, env vars
LangGraph	langgraph dep or StateGraph usage	Full — graphs, nodes, routes, tools, lint
AutoGen	dependency or AutoGen imports	Full — agents, teams, tools, runs, lint
Smolagents	dependency or Smolagents imports	Full — agents, managed agents, tools, runs, lint
Google ADK	google-adk dep	Full — agents, tools, hierarchy, workflows
OpenAI Agents SDK	openai-agents dep or agents imports	Full — agents, handoffs, tools, runners, lint
PydanticAI, LlamaIndex, Semantic Kernel, Agno	dependency or SDK imports	Dedicated agents, tools, teams/workflows and runs
LangChain, Haystack, Strands, CAMEL-AI	dependency, imports or factories	Dedicated agents, tools, workflows and invocations
MetaGPT, BeeAI, OpenAI Swarm	dependency or SDK imports	Dedicated roles/agents, teams and runs
Mastra, Vercel AI SDK Agents	npm dependency and agent API	JavaScript/TypeScript agents, tools, workflows and scripts
Semantic Kernel .NET/Java	project dependency or agent API	Agent initializers, instructions and invocations
Other Python or JS/TS frameworks	common constructors/factories	Generic agents, tools, relationships and entrypoints

For frameworks with full analysis, AgentLantern parses agents, tasks, tools, environment variables, and generates Mermaid diagrams.

The generic fallbacks cover conventional Python and JavaScript/TypeScript agent APIs. Additional analyzers can be installed through the agentlantern.analyzers entry-point interface.

Generated Files

File	Content
overview.md	Project snapshot, entrypoints, high-level flow
architecture.md	System map, key files, dependencies, environment variables
diagrams.md	Mermaid diagrams: agent-task graph, execution flow, sequence
agents.md	Agent roles, goals, tools, backstories
tasks.md	Task descriptions, agents assigned, expected outputs
runbook.md	Install, configure, run, static checks, troubleshooting
contact.md	Project contacts from pyproject.toml
index.html	Docsify site — publishable to GitHub Pages
_sidebar.md	Navigation menu
agentlantern-docs.html	Single-file HTML bundle; Mermaid rendering uses a CDN

Typical Workflow

# 1. Navigate to (or point at) your agent project
cd my-agent-project

# 2. Generate and view documentation
lantern web

# 3. Open http://localhost:9000
# 4. Press Ctrl+C to stop the server

Multiple Projects

# Terminal 1
lantern web ~/projects/crew-project --port 9000

# Terminal 2
lantern web ~/projects/adk-project --port 9001

Sharing Documentation

The generated docs/ folder is a static site. To share it:

• GitHub Pages — push docs/ and enable GitHub Pages in repository settings
• Single-file share — send docs/agentlantern-docs.html (no server; diagram rendering requires network access)
• Any web server — docs/ works on Nginx, Vercel, Netlify, etc.

Next Steps

• Examples
• FAQ