AgentShip

The runtime-agnostic production layer for AI agents.

AgentShip is LiteLLM for agent runtimes — one interface to run your agents on Google ADK, LangGraph, or any future engine, plus the production stack (API, sessions, streaming, observability, MCP) that those engines do not ship with.

Write your agent once. Swap runtimes with one YAML line. Ship in an hour, not two weeks.

Fast path (3 commands)

git clone https://github.com/Agent-Ship/agent-ship.git
cd agent-ship
make docker-setup   # API + PostgreSQL + Studio at http://localhost:7001

Then open http://localhost:7001/studio and start chatting with the built-in agent.

Where to start

• Just trying it out → Quick Start

• Deciding if AgentShip fits → Why AgentShip?

• About to build an agent → Building Agents Overview

• Integrating MCP tools → MCP Integration

Concepts in one screen

• Engines — ADK or LangGraph + LiteLLM; set execution_engine: per agent in YAML.

• Agents — two files per agent (YAML + Python) in src/all_agents/; auto-discovered at startup.

• Sessions — PostgreSQL-backed conversation history keyed by user_id:session_id.

• MCP tools — STDIO and HTTP/OAuth transports; tools auto-documented for the LLM.

• Observability — Opik tracing out of the box.

• Studio — browser UI at /studio for chatting, inspecting traces, and stopping runs.

---

Getting Started

• Why AgentShip?
• Architecture
• Quick Start
• Installation (Local Development)
• Configuration
• Docker Setup

Building Agents

• Building Agents
• Agent Configuration
• BaseAgent
• Tools
• Single Agent Pattern
• Orchestrator Pattern
• Tool Pattern

MCP Integration

• MCP Integration

Deployment

• Deployment
• Heroku Deployment

Testing

• Testing
• Writing Tests

Reference

• API Reference
• Contributing