ARES

ARES — a Go-native, multi-agent runtime with an adaptive knowledge graph

Architecture Modules Guides

ARES is a Go-native multi-agent runtime. It provides a unified SDK for building LLM-powered agents with built-in memory, tool calling, an adaptive knowledge graph (AKG), strategy evolution, and MCP integration.

Key features

  • Unified Runtime — one entry point (sdk.NewRuntime) wires the LLM client, tool registry, memory, knowledge fabric, and evolution system.
  • Adaptive Knowledge Graph (AKG) — distills conversations into fact-level KnowledgeObjects with quality gating and hybrid retrieval (vector + lexical), no LLM required in the build/retrieve loop.
  • Multi-provider LLM — OpenAI, Ollama, Anthropic, OpenRouter, plus automatic failover across fallback providers.
  • Tool calling — built-in tools, custom tools, MCP-discovered tools, and a capability planner for intent-based tool resolution.
  • Strategy evolution — GA-based optimization of agent instructions and LLM parameters, with a coordinator that applies patches from multiple sources.
  • Bilingual documentation — every module page is available in English and Chinese, kept in sync against the real source code.

Explore

Start with the Architecture overview, then browse the Module reference for per-package details, or read the Guides for deployment and extension walkthroughs.