Galileo opens agent controls
Galileo released an open-source toolset that gives platform teams finer-grained controls and visibility into multi-agent orchestration, aiming to close the observability gap for production agents reported. It’s pitched as a practical layer for debugging, policy enforcement, and workflow management across providers.
Galileo published Agent Control on March 11, 2026 and released the source code under an Apache‑2.0 license on GitHub. galileo.ai The control plane wires into agent code via a single @control() decorator to mark step‑level decision boundaries and enforces rules from a centralized policy server that supports hot‑reloadable controls without redeploying agents. galileo.ai Production install paths include a Docker Compose quickstart that brings up PostgreSQL plus a UI/dashboard and a health endpoint, and published SDKs for Python (agent-control‑sdk on PyPI) and TypeScript for immediate integration. github.com The repo and docs list built‑in, pluggable evaluators (regex, list, JSON, SQL) and framework adapters for LangChain, CrewAI, Google/OpenAI agent SDKs and AWS/Strands integrations, with early partner support named by The New Stack and vendor PRs including AWS, CrewAI, Glean, Strands Agents and Cisco AI Defense. github.com Galileo positions Agent Control as a response to analyst forecasts that agent deployment will scale rapidly—IDC’s FutureScape projects G2000 agent use rising tenfold by 2027 and token/API loads spiking ~1,000x—making centralized, runtime policy enforcement a capacity and governance requirement. idc.com Docs and the README call out audit‑ready traces, runtime configuration via API/UI, and the ability to reuse controls across agent fleets, features designed to cut the “one‑policy‑per‑agent” ticket proliferation that teams report when guardrails are hardcoded. docs.agentcontrol.dev
