Illuminate your multi-agent LLM systems directly in the editor.
AgentLantern is built for AI agent projects across frameworks. It detects the framework, reads project files statically, and gives you documentation, lint diagnostics, inspection, and editor feedback without calling any LLM.
CrewAI currently has the deepest support. Other agent frameworks are detected and documented as their analyzers are expanded.
Features
Real-time Lint Diagnostics
Errors and warnings appear in the Problems panel as you save files. No manual run needed.
CrewAI currently has the full lint rule set:
E001Task has no agentW006Missing required YAML fieldW007Unknown or misspelled fieldW008Wrong field type- …and 13 rules total
Code Lens
Where supported, AgentLantern shows one-line summaries above agent and task definitions. CrewAI crew.py files currently have the richest Code Lens support: role, tools, assigned tasks, and LLM without leaving the file.
🤖 Agent — Senior Data Analyst · tools: CSVSearchTool · tasks: analyze_data
def data_analyst(self) -> Agent:
Docs Preview
AgentLantern: Preview Docs regenerates the full documentation site and opens it in a VS Code side panel — overview, architecture, agents, tasks, diagrams, runbook.
Play and Replay
Launch the animated runtime viewer from VS Code:
AgentLantern: Play Live Runopens a terminal, runslantern play ., and launches the Play UI.AgentLantern: Replay Saved Runprompts for a replay name or.jsonlpath and runslantern replay.
These commands run in a VS Code terminal because Play and Replay are long-running local servers. Stop them with the UI STOP button or Ctrl+C in the terminal.
Command Palette
| Command | What it does |
|---|---|
AgentLantern: Lint Project |
Run lint → Problems panel |
AgentLantern: Generate Docs |
Write docs to docs/ |
AgentLantern: Preview Docs |
Generate + open WebView |
AgentLantern: Play Live Run |
Start lantern play in a VS Code terminal |
AgentLantern: Replay Saved Run |
Start lantern replay in a VS Code terminal |
AgentLantern: Inspect Project (JSON) |
Open raw project model |
Requirements
- VS Code 1.85+
- Python 3.11+
- AgentLantern CLI installed:
pip install agentlantern
# or (recommended)
uv tool install agentlantern
Configuration
| Setting | Default | Description |
|---|---|---|
agentlantern.lanternPath |
"lantern" |
Path to the lantern binary |
agentlantern.lintOnSave |
true |
Auto-lint on Python / YAML save |
agentlantern.codeLensEnabled |
true |
Show Code Lens where supported; strongest today for CrewAI crew.py |
Supported Frameworks
| Framework | Detection | Docs | Lint | Code Lens |
|---|---|---|---|---|
| CrewAI | ✅ | ✅ Full | ✅ Full | ✅ agents + tasks |
| Google ADK | ✅ | ✅ Full | Planned | Planned |
| LangGraph | ✅ | Basic | Planned | Planned |
| AutoGen | ✅ | Basic | Planned | Planned |
| Smolagents | ✅ | Basic | Planned | Planned |
Troubleshooting
Diagnostics not appearing — verify lantern is on your PATH:
# In VS Code terminal
lantern --version
If not found, set agentlantern.lanternPath to the full binary path.
Code Lens not showing — ensure agentlantern.codeLensEnabled is true. CrewAI Code Lens currently expects the file to be named crew.py.