Show HN: Kodosumi – Open-source runtime for AI agents (www.kodosumi.io)

Kodosumi is an open-source, pre‑configured runtime for building, deploying and scaling AI agents built on Ray (for distributed execution), FastAPI (agent endpoints) and Litestar (admin/core services). It packages a developer workflow into a minimal YAML + config.py configuration, includes real‑time monitoring (Ray dashboard + built‑in web panel with event streaming and replay), and supports deployment across Docker, Kubernetes or bare metal. You can install it via pip, start a local Ray cluster, deploy services with Ray Serve and register endpoints with the koco CLI—no deep Ray expertise required if you already code agents in Python. For the AI/ML community this lowers the barrier to productionizing agentic systems: Kodosumi targets long‑running workflows, bursty traffic and complex multi‑agent flows by leveraging Ray’s horizontal scaling and scheduling, while remaining framework‑agnostic (any LLMs, vector stores, SDKs or self‑hosted models). It avoids vendor lock‑in and offers an extensible “agent → flow → agentic service” model for composing services. Caveats: the project is under active development (some APIs/concepts may change), but it builds on enterprise‑grade components and links into a marketplace/eco system (Masumi/Sokosumi) for monetization and community support.