Curie: Open-Source Agent Deployment Platform Bridges Local-to-Production Gap
Key Takeaways
- ▸Open-source, Apache 2.0 licensed platform eliminates environment-driven agent failures through a three-tier deployment parity ladder
- ▸Supports deployment across laptops, Docker Compose, and Kubernetes clusters with consistent behavior via immutable plugin bundles
- ▸Integrates with Claude Code plugin format and supports multiple LLM providers including Anthropic, OpenRouter, and local Ollama instances
Summary
Anthropic has released Curie, an open-source (Apache 2.0) deployment platform for production AI agents built with Claude Code. Curie enables developers to author agent plugin bundles (skills, tools, and MCP servers) and deploy them across multiple environments—from local development to Kubernetes clusters—while maintaining consistent behavior and catching environment differences early.
The platform introduces a three-tier deployment ladder: "skill" runs agents directly in a container without platform overhead, "local" runs through a full platform via Docker Compose for testing, and "cluster" deploys to production Kubernetes. This design ensures the same immutable plugin bundle works identically across all three tiers, surfacing environment discrepancies during development rather than after users encounter them.
Curie supports multiple LLM providers (Anthropic, OpenRouter, and self-hosted via Ollama), features git-driven deployments, and includes built-in tracing, evals, and budget tracking. Initial Slack integration is available, with email and Teams support coming next. The platform is self-hostable and fully open source, with a CLI tool (curie) orchestrating the entire workflow from authoring to deployment.
- Includes built-in traces, evals, budget controls, and git-driven deployments with no additional cost
- Initial Slack bot support with email and Teams integration roadmapped
Editorial Opinion
Curie directly addresses a critical pain point in agent deployment: the gap between local development and production that causes seemingly working agents to fail when scaled. By enforcing environment parity through its three-tier architecture and immutable bundles, it shifts environment bugs from user-facing production failures to the development phase where they're cheap to fix. For teams shipping Claude Code agents, this open-source approach could significantly accelerate the journey from prototype to reliable production service.
