trpc-group/trpc-agent-go

Add via Go module: go get trpc.group/trpc-go/trpc-agent-go
Clone and run the example: git clone https://github.com/trpc-group/trpc-agent-go.git && cd trpc-agent-go && export OPENAI_API_KEY=... && cd examples/runner && go run . -model="gpt-4o-mini" -streaming=true

Usage

• •Create a model and agent: openai.New("deepseek-chat", ...) + llmagent.New("assistant", llmagent.WithModel(...), llmagent.WithTools(...))
• •Run via Runner with session/memory: runner.NewRunner("app", agent).Run(ctx, userID, sessionID, message)
• •Compose multi-agent workflows: chainagent.New / parallelagent.New / CycleAgent
• •Add MCP tools: mcpTool := mcptool.New(serverConn); expose frontends via AG-UI (SSE) and interop via A2A

Key features

• ✓GraphAgent: type-safe graph workflows with multi-conditional fan-out routing, functionally equivalent to LangGraph for Go
• ✓Multi-agent orchestration (chain, parallel, cycle) plus planner + executor loop
• ✓Rich tool ecosystem: function tools, MCP tools (via trpc-mcp-go), web search, code execution, file tools
• ✓Persistent session/memory (in-memory or Redis), RAG knowledge retrieval, versioned artifacts (in-memory, S3, COS)
• ✓Agent Skills (SKILL.md spec) with safe execution, agent self-evolution (Hermes-style session reviews), prompt caching (~90% savings), evaluation & benchmarks, OpenTelemetry + Langfuse observability

Best for

Go services that need concurrent, observable, production-grade agent applications integrating with MCP tools and AG-UI/A2A frontends.

Caveats

• ⚠Requires Go 1.21+ and an LLM provider API key (OpenAI, DeepSeek, etc.)
• ⚠Do not break the event loop when cancelling — always cancel the context and keep draining the event channel
• ⚠When using WithCodeExecutor only for skill_run, disable the response code-execution processor to avoid auto-executing fenced code blocks (2.x guidance)