tRPC-Agent-Go is a Go framework from Tencent’s tRPC group for building production-grade AI agent systems. It provides graph-based workflows, tool use, memory, multi-agent orchestration and first-class support for the A2A and MCP protocols, with evaluation and OpenTelemetry observability built in. It is an open-source project on GitHub with around 1,900 stars and very active development through 2026.
Most agent frameworks are Python-first; tRPC-Agent-Go brings the same capabilities to Go services that need agent features inside existing high-performance backends. Agents are assembled from reusable runners, models, tools and knowledge components, and can be wired into graph workflows for deterministic control over multi-step processes. The framework speaks the Agent-to-Agent (A2A) protocol for cross-agent collaboration, consumes MCP servers for tools, supports AG-UI for streaming agent interfaces, and includes RAG support. Built-in evaluation harnesses and tracing make it suitable for teams shipping agents into real production traffic rather than demos.
Free. tRPC-Agent-Go is open source under the Apache-2.0 license; you supply your own model API keys.
The framework is model-agnostic, plugging into major hosted LLM APIs and local inference endpoints through a common model interface, and it manages agent sessions and state so long-running, multi-turn agents behave predictably under concurrency. Example projects in the repository walk through single agents, graph workflows and multi-agent teams, and because it grows out of Tencent’s tRPC ecosystem, it inherits conventions that backend teams already using tRPC-Go will find familiar, including configuration and service deployment patterns.
Go engineering teams — especially those already on tRPC or similar microservice stacks — who want to add reliable, observable LLM agents to production systems without adopting a Python sidecar.