Orpheus

[](https://go.dev/) [](https://www.typescriptlang.org/) [](LICENSE)

Orpheus is an Agent Runtime that solves the 'Silent Failure' of CPU autoscaling. It scales on queue depth, persists memory, and recovers from crashes automatically.

It uses queue-depth autoscaling and lightweight runc containers to handle agent workloads that traditional serverless platforms can't support.

Built for agents that call LLMs, orchestrate multi-step workflows, and maintain context across invocations, Orpheus provides the execution layer with persistent workspaces, built-in streaming, and production-grade supervision.

---

**Key capabilities:**

• Sub-second container startup using runc instead of Docker
• Queue-depth autoscaling that scales on pending work, not CPU metrics
• Persistent workspaces that survive container restarts and worker recycling
• MCP endpoints — Every deployed agent is automatically discoverable via Model Context Protocol
• Built-in SSE streaming for real-time agent output
• vLLM/Ollama integration with automatic lifecycle management
• Production supervision with circuit breakers, backoff, and OOM handling
• Prometheus metrics — Built-in /metrics for queue depth, workers, and executions

---

Getting Started

Quick Start (Recommended)

One command sets up everything:

git clone https://github.com/arpitnath/orpheu