trpc-group/trpc-agent-go

What it solves

tRPC-Agent-Go is a Go-native framework designed for building production-ready AI agent systems. It addresses the complexity of moving from simple LLM prompts to scalable, observable, and controllable agentic workflows that integrate seamlessly into existing Go services.

How it works

The framework provides a comprehensive stack for agent orchestration, including:

• Agent Runtimes: Streaming runners that support context cancellation and service-friendly APIs.
• Orchestration Patterns: Support for linear chains, parallel execution, and cyclic workflows. It also includes a GraphAgent for type-safe, conditional routing and complex state management (similar to LangGraph).
• Tooling & Integration: A rich ecosystem including function tools, Model Context Protocol (MCP) support, web search, and code execution.
• State Management: Persistent session and memory services (including Redis support) and RAG-powered knowledge retrieval.
• Self-Evolution: A system that reviews completed sessions to extract and publish reusable workflows as "Agent Skills" (SKILL.md).
• Observability: Built-in OpenTelemetry tracing and metrics, with examples for Langfuse integration.

Who it’s for

Developers and engineers building enterprise AI applications in Go, such as customer support bots, data analysis assistants, DevOps automation tools, and RAG-powered knowledge management systems.

Highlights

• Go-Native Performance: Leverages Go's concurrency and streaming capabilities for efficient agent execution.
• Graph-Based Workflows: Type-safe graph agents with multi-conditional routing for complex logic.
• passage to Production: Integrated OpenTelemetry and evaluation benchmarks to measure quality over time.
• Agent Self-Evolution: Ability to automatically promote successful session patterns into reusable skills.
• MCP Support: Integration with the Model Context Protocol for standardized tool usage.
• Prompt Caching: Automatic cost optimization to reduce LLM expenses.