Show HN: Hector – Pure A2A-Native Declarative AI Agent Platform (Go) (github.com)

Hector is an alpha-stage, Go-native platform that lets you define full AI agents entirely in YAML and serve them over the A2A (Agent-to-Agent) protocol — no Python glue or handwritten orchestration required. It emphasizes declarative agent definitions (slot-based prompts, reasoning instructions, LLM configs), built-in RAG (semantic search/document stores), streaming token output, and an “agent_call” tool that enables LLM-driven delegation and multi-agent orchestration. Because it’s A2A-compliant, Hector can host native agents, discover and call remote agents by URL, and publish capability cards for cross-organization interoperability. Technically, Hector supports multiple reasoning engines (chain-of-thought and a supervisor mode optimized for delegation), plugin extensibility via gRPC, integrations with LLM backends (OpenAI, Anthropic examples provided), document stores (Qdrant), and MCP connectors to many apps. It includes production-focused features — JWT/OAuth2/OIDC auth, visibility controls, sessions, streaming via WebSocket, command/file tool whitelisting and sandboxing, logging/error handling — and ships as a Go library, CLI, and Docker image. The key implication: teams can rapidly compose, share, and orchestrate purpose-built agents across ecosystems using a standardized protocol and purely declarative configs, lowering engineering friction compared with code-first frameworks like LangChain or AutoGen.