agno-agi/agent-platform-railway

agent-platform-railway is a starter template for building AI agents, built on top of Agno, an open source Python framework for creating agent systems, and designed to be deployed to the Railway hosting platform with a single command. The project runs entirely in your own cloud account with your own authentication and your own database, rather than depending on any third party service beyond the AI model you choose. What makes this template distinctive is that it is built specifically to be developed and maintained by AI coding assistants like Claude Code, not just by a human developer. Because agent code, logs, trace data, and the tools used to iterate on the system all live together in one codebase, a coding assistant can read the whole system and make changes to it directly. The README describes five workflows for this: creating a new agent from a short back and forth with the assistant, improving an existing agent by having the assistant test it against its own written instructions and fix failures, extending an agent with new features under your direction, running an evaluation suite and fixing whatever fails, and sweeping the repository for places where documentation, code, and configuration have drifted out of sync. Under the hood, the platform runs on FastAPI serving Agno's AgentOS runtime, stores sessions, memory, and traces in PostgreSQL with the pgvector extension for embeddings, and can connect agents to many external tools and MCP servers through Agno's toolkit system. Agents can be exposed through Slack, with Discord and Telegram support available as add-ons. Locally the system runs in Docker, and in production it deploys to Railway. Getting started means cloning the repository, setting an OpenAI API key in a local .env file, and running the stack with Docker Compose, after which you can connect a chat interface at os.agno.com and start creating and testing agents. The project also ships an automated evaluation suite that runs cases through Agno's judge and reliability checks, with results stored in Postgres so you can track how agents perform over time. Deploying to production uses Railway's CLI and a separate production environment file, and the README notes the first production deploy is expected to fail by design because token based authorization is enabled by default until configured.