AgenticGoKit

🚀 BETA RELEASE - The v1beta API is now stable and recommended for all new projects. While still in beta, the core APIs are working well and ready for testing. We continue to refine features and welcome feedback and contributions!

📋 API Versioning Plan:

• Current (v0.x): v1beta package is the recommended API (formerly vnext)
• v1.0 Release: v1beta will become the primary v1 package
• Legacy APIs: Both core and core/vnext packages will be removed in v1.0

Robust Go framework for building intelligent multi-agent AI systems

The most productive way to build AI agents in Go. AgenticGoKit provides a unified, streaming-first API for creating intelligent agents with built-in workflow orchestration, tool integration, and memory management. Start with simple single agents and scale to complex multi-agent workflows.

• v1beta APIs: Modern, streaming-first agent interface with comprehensive error handling
• Multimodal Support: Native support for images, audio, and video inputs alongside text
• Real-time Streaming: Watch your agents think and respond in real-time
• Multi-Agent Workflows: Sequential, parallel, DAG, and loop orchestration patterns
• Production-Ready Observability: Built-in distributed tracing with OpenTelemetry support
• Multiple LLM Providers: Seamlessly switch between OpenAI, Anthropic, Ollama, Azure OpenAI, Azure AI Foundry Local, HuggingFace, OpenRouter, vLLM, BentoML, MLFlow, and more
• High Performance: Compiled Go binaries with minimal overhead
• Batteries Included: Built-in memory and RAG by default (zero config needed, swappable with pgvector/custom)
• Rich Integrations: Memory providers, tool discovery, MCP protocol support
• Active Development: Beta status with stable core APIs and ongoing improvements

AgenticGoKit is one part of a three-part ecosystem designed to take you from idea to production:

Typical flow: Design with the core framework → Scaffold with AGK using official templates → Build & ship with your preferred deployment stack.

📦 Explore the full ecosystem at the AgenticGoKit Organization.

Start building immediately with the modern v1beta API:

package main

import (
    "context"
    "fmt"
    "log"
    "time"
    
    "github.com/agenticgokit/agenticgokit/v1beta"
)

func main() {
    // Create a chat agent with Ollama
    agent, err := v1beta.NewBuilder("ChatAgent").
        WithConfig(&v1beta.Config{
            Name:         "ChatAgent",
            SystemPrompt: "You are a helpful assistant",
            LLM: v1beta.LLMConfig{
                Provider: "ollama",
                Model:    "gemma3:1b",
                BaseURL:  "http://localhost:11434",
            },
        }).
        Build()
    if err != nil {
        log.Fatal(err)
    }

    // Basic execution
    result, err := agent.Run(context.Background(), "Explain Go channels in 50 words")
    if err != nil {
        log.Fatal(err)
    }
    
    fmt.Println("Response:", result.Content)
}

Enable observability with a single environment variable:

export AGK_TRACE=true  # Automatic tracing to .agk/runs/<run-id>/trace.jsonl

Note: CLI tooling for AgenticGoKit is provided by the agk package. Install with: go install github.com/agenticgokit/agk@latest

AgenticGoKit handles the complexities of building AI systems so you can focus on logic.

Orchestrate multiple agents using robust patterns. Pass data between agents, handle errors, and manage state automatically.

• Patterns: Sequential, Parallel, DAG, Loop, and SubWorkflows.
• Visuals: Auto-generate Mermaid diagrams for your flows.
• Example: Sequential Workflow Demo

Built from the ground up for streaming. Receive tokens and tool updates as they happen, suitable for real-time UI experiences.

• Example: Streaming Workflow

Batteries Included: Agents come with valid memory out-of-the-box (chromem embedded vector DB).

• Features: Chat history preservation, semantic search, and document ingestion.
• Configurable: Swap the default with pgvector or custom providers easily.

Native support for Images, Audio, and Video inputs. Works seamlessly with models like GPT-4 Vision, Gemini Pro Vision, etc.

Extend agents with tools using standard Go functions or the Model Context Protocol (MCP).

• Dynamic Discovery: Automatically find and register tools from MCP servers.
• Standardized: Support for the emerging standard for LLM tool interoperability.

Production-Ready: Built-in distributed tracing with zero configuration required.

• Features: OpenTelemetry integration, workflow trace hierarchies, OTLP/Jaeger support.
• Exporters: Console, file, and OTLP for complete visibility into agent execution.
• Example: Observability Basics

AgenticGoKit works with all major LLM providers out of the box:

• Getting Started - Build your first agent
• API Reference - Comprehensive API docs
• Observability Guide - Distributed tracing and monitoring
• Memory & RAG - Deep dive into memory systems

• Story Writer Chat v2 - Complete Real-time collaborative writing app
• Ollama Quickstart - Local LLM development
• Foundry Local Quickstart - On-device inference with Azure AI Foundry Local
• MCP Integration - Using Model Context Protocol
• HuggingFace Quickstart - Using HF Inference Endpoints
• BentoML Quickstart - Self-hosted ML models
• MLFlow Gateway Demo - MLFlow AI Gateway integration
• vLLM Quickstart - High-throughput inference

• Recommended: Use v1beta package for all new projects
• Import Path: github.com/agenticgokit/agenticgokit/v1beta
• Stability: Beta - Core APIs are stable and functional, suitable for testing and development
• Status: Beta - Core APIs are stable; v1beta is the evolution of the former core/vnext package
• Note: core package is legacy and will be removed in v1.0

What's Changing:

• v1beta package will become the primary v1 API
• Legacy core and core/vnext packages will be removed entirely
• Clean, stable API with semantic versioning guarantees

Migration Path:

• If you're using v1beta or vnext: Minimal changes (import path update only)
• If you're using core: Migrate to v1beta now to prepare
• core/vnext users: vnext has been renamed to v1beta - update imports

Timeline:

• v0.x (Current): v1beta stabilization and testing
• v1.0 (Planned): v1beta → v1, remove core package

The v1beta package represents our next-generation API design:

• ✅ Streaming-first architecture
• ✅ Unified builder pattern
• ✅ Better error handling
• ✅ Workflow composition
• ✅ Stable core APIs (beta status)
• ⚠️ Minor changes possible before v1.0

By using v1beta today, you're getting access to the latest features and helping shape the v1.0 release with your feedback.

• Website: www.agenticgokit.com
• Documentation: docs.agenticgokit.com
• Examples: examples/
• AGK CLI: github.com/AgenticGoKit/agk
• Template Registry: github.com/agk-templates
• Discussions: GitHub Discussions
• Issues: GitHub Issues
• Discord: Join the Community

We welcome contributions! See docs/contributors/ContributorGuide.md for getting started.

Apache 2.0 - see LICENSE for details.