AgentVoy — Universal AI Agent Development Platform<br>v0.5.0 — DevTools, one-command deploy, 7 frameworks<br>Build and deploy<br>agentic apps.<br>Single agents or multi-agent pipelines — with real-time DevTools, Streamlit chat UI, and one-command deploy to Docker or Fly.io. Any framework. Any model.<br>Live DevToolsOne-command deploy7 frameworksMulti-agent pipelinesFastAPI + StreamlitDocker · Fly.io · AWS · GCP<br>$npx agentvoy create my-project<br>Create project →<br>GitHub →npm →

Chat + DevTools — split view

Deploy your agentic app to

Docker

Fly.io

Railway

GCP

AWS

Frameworks<br>One CLI. Any framework.<br>Pick your stack. AgentVoy generates the right boilerplate for each framework automatically.<br>OpenAI Agents SDK<br>Python<br>●Available

Google ADK<br>Python<br>●Available

CrewAI<br>Python<br>●Available

LangGraph<br>Python<br>●Available

Anthropic SDK<br>Python<br>●Available

LlamaIndex<br>Python<br>●Available

AutoGen<br>Python<br>●Available

Model Providers<br>Any model. Bring your key.<br>OpenAI, Anthropic, Google, Groq, Mistral — or run fully local with Ollama. No GPU required.<br>OpenAI<br>GPT-4o, o1

Anthropic<br>Claude Sonnet 4, Opus 4

Google<br>Gemini 2.0 Flash

Ollama<br>llama3, mistral (local)

Groq<br>llama-3.3-70b

Mistral<br>mistral-large

DevTools<br>See your agent think.<br>Every app-mode project includes a real-time DevTools dashboard. Start with agentvoy dev and open /dev.

Creating a project with Anthropic SDK

agentvoy dev<br>Starts your agent server with hot-reload, opens the DevTools dashboard, and streams trace events via WebSocket. Zero config needed.<br>$ agentvoy dev<br>API Endpoints<br>GET/dev—DevTools dashboard<br>WS/ws/trace—Real-time trace stream<br>GET/dev/events—All events as JSON<br>POST/run—Run your agent<br>GET/health—Health check

Dynamic model switching<br>The Streamlit chat UI auto-detects API keys from .env and shows available models — GPT-4o, Claude Sonnet, Gemini Flash, and more. Switch models on the fly without restarting.

agent.guard.yml<br>Guardrails built in,<br>not bolted on.<br>Every AgentVoy project ships with agent.guard.yml — a universal declarative config for permissions, cost limits, and behavior constraints. One format that works across every framework. Enforced at runtime by agentvoy-guard — automatically included in every project.<br>🔒Network permissions<br>Allowlist/denylist domains your agent can reach

💰Cost limits<br>Hard cap spend per run — never get a surprise bill

🔁Iteration caps<br>Set max tool calls and loop iterations

🛡️Prompt injection blocking<br>Detect and block prompt injection attacks

👁️PII detection<br>Warn or block when sensitive data is present

agent.guard.yml<br># Universal guardrails — works across all frameworks<br>version: "1.0"

model:<br>provider: anthropic<br>model: claude-sonnet-4-20250514

permissions:<br>network:<br>mode: restricted<br>allow: ["*.github.com"]<br>execution:<br>allow_shell: false

guardrails:<br>input:<br>block_prompt_injection: true<br>pii_detection: warn<br>behavior:<br>max_iterations: 20<br>cost_limit: "$1.00"<br>timeout: 5m

Why AgentVoy<br>Everything you need to ship agents.<br>From zero to production-ready agent project in one command.<br>Two paths, one command<br>Build a local agent for experimentation, or a full deployable app with FastAPI server, Streamlit chat UI, DevTools dashboard, and cloud configs — all from npx agentvoy create.

🔬<br>Live DevTools<br>Real-time agent tracing via WebSocket. See every LLM call, tool invocation, guard check, and pipeline stage as it happens — with a dark-themed dashboard at /dev.

🚀<br>One-command deploy<br>agentvoy deploy --target docker builds and runs your container. agentvoy deploy --target fly-io deploys to the cloud. Secrets, configs, and health checks handled automatically.

🔒<br>Secure by default<br>Every project ships with agent.guard.yml. Permissions, cost caps, iteration limits, prompt injection blocking — enforced before you write a line of agent logic.

🔌<br>Any framework. Any model.<br>OpenAI, Google ADK, CrewAI, LangGraph, Anthropic, LlamaIndex, AutoGen. Switch models on the fly from the chat UI — no restart needed.

🤖<br>Multi-agent pipelines<br>Build sequential pipelines where each agent builds on the previous stage. Name your agents, and get a production-ready pipeline with real-time stage visualization.

Deploy<br>One command. Deploy anywhere.<br>agentvoy deploy builds, configures, and ships your agent. Your agent.guard.yml guardrails flow directly into infrastructure config.<br>agent.guard.yml → cloud config<br>behavior.timeout: 5m→Docker HEALTHCHECK, Cloud Run timeout<br>behavior.cost_limit: $1.00→Container memory: 512Mi<br>execution.allow_shell: false→Non-root Docker user

🐳Docker<br>Builds the image, runs the container, exposes API + DevTools<br>+ Dockerfile<br>+ .dockerignore<br>+ docker-compose.yml

$ agentvoy deploy --target docker<br>✈️Fly.io<br>Checks auth, sets secrets from .env, deploys via flyctl<br>+ deploy/fly.toml

$ agentvoy deploy --target fly-io<br>🚂Railway<br>Generates railway.json config for one-click deploy<br>+ deploy/railway.json

$ agentvoy deploy --target railway<br>☁️GCP Cloud Run<br>Generates Cloud Run service config with health checks<br>+ deploy/cloud-run.yaml

$ agentvoy deploy --target...