AgentVoy just dropped as a CLI tool that promises to solve the biggest pain point in AI agent development: the endless boilerplate setup before you can even start building something useful. Created by Chinmay Murugkar and published on GitHub, this thing scaffolds production-ready agents across seven major frameworks in seconds—one command, any framework, deploy anywhere.

Two Paths, One Goal

The tool splits into two modes upfront. "Agent" mode gives you a flat project structure for fast local dev with an interactive REPL and zero infrastructure headaches—perfect for experimentation. "App" mode goes full production: FastAPI server, Streamlit chat UI, real-time DevTools dashboard, Docker configs, and cloud deployment files all generated automatically. You can even spin up multi-agent pipelines (researcher → writer → reviewer) with a single flag.

Seven Frameworks, One Config

AgentVoy supports OpenAI Agents SDK, Google ADK, CrewAI, LangGraph, Anthropic SDK, LlamaIndex, and AutoGen—all Python-based. It integrates with providers like OpenAI (GPT-4o, o1), Anthropic (Claude Sonnet 4, Opus 4), Google (Gemini 2.0 Flash, 2.5 Pro), Ollama for local models, Groq, and Mistral. The selling point: you write your agent logic once, swap frameworks via CLI flags, and deploy without rewriting anything.

Built-In DevTools That Actually Work

Every app-mode project includes a dark-themed observability dashboard at /dev with WebSocket-powered real-time trace streaming. You get the full event timeline—agent_start, llm_call, tool_call, guard_check, pipeline_stage, agent_complete—and can click any event to inspect model details, token counts, latency, and tool I/O. No extra dependencies, just a single-page HTML dashboard.

Guardrails as Code

Every project ships with agent.guard.yml, a declarative config that handles permissions (network access whitelists, filesystem read/write restrictions, shell/subprocess blocking) and guardrails (prompt injection detection, PII filtering, content moderation, cost limits, iteration caps). The runtime enforcement library—agentvoy-guard—wraps your agent session and checks inputs/outputs against these policies automatically. Security isn't bolted on; it's baked into the scaffolding.

One-Command Deployment

Deploy targets include Docker (generates Dockerfile + compose), Fly.io (one-step with flyctl auth), Railway, GCP Cloud Run, and AWS Lambda. Guard config maps directly to infrastructure settings—timeout values become container healthcheck intervals, cost limits set memory allocations. You can even deploy an existing agent project without touching your code: it generates server.py, streamlit_app.py, devtools.html, and all deployment files around your logic.

Key Takeaways

  • Seven framework adapters out of the box (OpenAI, Google ADK, CrewAI, LangGraph, Anthropic SDK, LlamaIndex, AutoGen)
  • DevTools dashboard with real-time WebSocket tracing included in every app-mode project
  • Guardrails codified in YAML flow into deployment configuration automatically
  • One-command deploy to Docker, Fly.io, Railway, GCP Cloud Run, or AWS Lambda

The Bottom Line

AgentVoy isn't trying to be clever—it's solving the boring problem that kills developer velocity. Boilerplate is the enemy of experimentation, and this tool removes it entirely for agent development specifically. Worth bookmarking if you're building anything beyond a single-agent demo.