Scale AI Open Sourced Agentex (github.com)

Scale AI has open‑sourced Agentex, a lightweight, cloud‑agnostic framework for building, running and scaling AI agents across five capability levels (L1–L5). The release includes an open source server, developer UI, Python SDK/CLI and local dev tooling so teams can prototype synchronous chat agents (L1–L3) and incrementally upgrade to asynchronous, long‑running or fully agentic workflows (L4–L5) without rearchitecting. The repo provides quickstart scaffolding (agentex init), a streaming example (acp.py), a manifest.yaml-driven runtime, a dev UI on localhost:3000, and notebook support; it’s Kubernetes‑native and integrates with Temporal for durable execution of complex, multi‑step flows. For practitioners the practical implications are immediate: Agentex supports multiple LLM providers via environment API keys, streaming and async agents, multi‑agent interactions via an Agent Developer Kit, and per‑agent hosting/traceability for debugging. Local prerequisites include Python 3.12+, the uv package manager, Docker/Node and handling Redis port conflicts; lazydocker is recommended for inspecting service health. The open source edition focuses on local development and community support, while Scale’s Enterprise offering layers in GitOps deployment, managed AgentOps (hosting, versioning, evaluation), identity/SSO and SLAs — making Agentex useful both for experimentation and as a path to production-grade, enterprise agent deployments.