Zymi Core

•
10
•

Provides an auditable backend for AI agents, managing tools as declarative, event-sourced, and approval-gated YAML pipelines.

Zymi Core is engineered as the robust backend for AI agents, specializing in the management and execution of tools and workflows rather than agent frontend logic. It enables the definition of agent capabilities as declarative YAML pipelines, which are inherently event-sourced to ensure complete auditability, replayability, and resumability of every action. This design prioritizes boundary safety, routing potentially risky operations through policy, contracts, and interactive human approval, often directly within the calling agent's UI via the Model Context Protocol (MCP). Zymi Core also facilitates powerful self-debugging, allowing agents to introspect their own runs and diagnose failures using detailed event traces. It can operate as an MCP server, exposing these governed pipelines as standardized tools to any MCP-compatible agent framework, or function standalone for various automation scenarios.

주요 기능

01Define agent tools and workflows using declarative YAML pipelines.
02Interactive human approval for critical agent actions via MCP elicitation in the caller's UI.
03Acts as an MCP server, exposing governed pipelines as standard tools to any compatible agent.
04Event-sourced execution for auditable, replayable, and resumable pipeline runs.
05Self-debuggable with observability tools to introspect runs and analyze event traces.
06AI-assistant friendly, generating documentation for improved project understanding and YAML generation.
079 GitHub stars

사용 사례

01Integrating complex, multi-step declarative workflows as single tools into existing agent frameworks like LangGraph or CrewAI.
02Building auditable AI agents with transparent and governed tool execution.
03Creating standalone, event-sourced automation agents, such as Telegram chatbots, with built-in approval flows.