Kunal Kushwaha

AgenticGoKit ❤️ HuggingFace and OpenRouter

· AgenticGoKit Team

AgenticGoKit Loves HuggingFace and OpenRouter

We’re excited to announce the release of AgenticGoKit v0.4.6 — a major update that expands how Go developers can build and experiment with agentic workflows.

With this release, AgenticGoKit now supports:

  • Hugging Face — Access models through the new router API with OpenAI-compatible format
  • OpenRouter — Use a wide variety of LLMs (GPT, Claude, Gemini, Llama) through a unified API

These integrations empower developers to build richer, more flexible AI agents without switching languages or managing complex SDKs.

What’s New

Hugging Face Integration with New Router API

AgenticGoKit now supports Hugging Face’s latest router-based architecture (router.huggingface.co), which provides:

  • OpenAI-compatible format — Seamless migration from other providers
  • Multiple API types — Inference API, Chat API, dedicated Endpoints, and self-hosted TGI
  • Llama 3.2 models — Access to the latest open-source models
  • Streaming support — Real-time token-by-token responses
  • Advanced parameters — Fine-tune with temperature, top-p, top-k, and more

Quick Example

import (
    "context"
    "github.com/kunalkushwaha/agenticgokit/core/vnext"
    _ "github.com/kunalkushwaha/agenticgokit/plugins/llm/huggingface"
)

config := &vnext.Config{
    Name:         "hf-assistant",
    SystemPrompt: "You are a helpful AI assistant.",
    LLM: vnext.LLMConfig{
        Provider:    "huggingface",
        Model:       "meta-llama/Llama-3.2-1B-Instruct",
        APIKey:      os.Getenv("HUGGINGFACE_API_KEY"),
        Temperature: 0.7,
        MaxTokens:   500,
    },
}

agent, _ := vnext.NewBuilder("hf-assistant").WithConfig(config).Build()
result, _ := agent.Run(context.Background(), "Explain machine learning")
fmt.Println(result.Content)

Available Models:

  • meta-llama/Llama-3.2-1B-Instruct — Fast, efficient
  • meta-llama/Llama-3.2-3B-Instruct — Better quality
  • deepseek-ai/DeepSeek-R1 — Advanced reasoning
  • Model routing with :fastest and :cheapest suffixes

HuggingFace Examples | HuggingFace vNext Quickstart

OpenRouter Integration

OpenRouter provides a single interface to dozens of leading LLMs from OpenAI, Anthropic, Google, Meta, and more. With AgenticGoKit v0.4.6, you can:

  • Access GPT-4, Claude 3, Gemini 2.0, Llama 3.1, and many others
  • Compare models side-by-side for your use case
  • Switch between providers with a single line of code
  • Track usage with site analytics

Quick Example

config := &vnext.Config{
    Name:         "openrouter-agent",
    SystemPrompt: "You are a helpful assistant.",
    LLM: vnext.LLMConfig{
        Provider:    "openrouter",
        Model:       "anthropic/claude-3-haiku",
        APIKey:      os.Getenv("OPENROUTER_API_KEY"),
        Temperature: 0.7,
        MaxTokens:   500,
    },
}

agent, _ := vnext.NewBuilder("openrouter-agent").WithConfig(config).Build()
result, _ := agent.Run(ctx, "What's the difference between REST and GraphQL?")

Popular Models Available:

  • openai/gpt-4-turbo — Most capable GPT model
  • anthropic/claude-3-haiku — Fast, cost-effective
  • google/gemini-2.0-flash-exp:free — Free Google model
  • meta-llama/llama-3.1-8b-instruct — Open-source Llama

OpenRouter Examples | OpenRouter vNext Quickstart

Why This Matters

Building agentic systems in Go is now easier and more powerful than ever. By connecting to Hugging Face and OpenRouter, AgenticGoKit developers can:

  • Rapidly prototype with different LLMs and compare performance
  • Optimize costs by testing various model/provider combinations
  • Build multi-agent systems using diverse model capabilities
  • Access the latest models including Llama 3.2, Claude 3, Gemini 2.0
  • Keep the speed and reliability of Go with type safety
  • Switch providers easily without rewriting application logic

Key Features in v0.4.6

Multiple API Types for HuggingFace

  1. Inference API — New router with OpenAI-compatible format
  2. Chat API — Conversational models with legacy support
  3. Inference Endpoints — Dedicated hosting for production
  4. Text Generation Inference (TGI) — Self-hosted optimized inference

Streaming Support

Both HuggingFace and OpenRouter now support real-time streaming:

stream, _ := agent.RunStream(ctx, "Write a story about AI")
for chunk := range stream.Chunks() {
    if chunk.Type == vnext.ChunkTypeDelta {
        fmt.Print(chunk.Delta)
    }
}

Advanced Configuration

Fine-tune model behavior with provider-specific parameters:

LLM: vnext.LLMConfig{
    Provider:          "huggingface",
    Model:            "meta-llama/Llama-3.2-1B-Instruct",
    Temperature:      0.8,
    MaxTokens:        1000,
    TopP:             0.9,
    TopK:             50,
    RepetitionPenalty: 1.2,
}

Comprehensive Examples

  • 9 HuggingFace examples covering all API types
  • 8 OpenRouter examples demonstrating various models
  • Simple quickstart examples for both providers
  • Migration guides for the new HuggingFace router API
  • Complete documentation with troubleshooting tips

The Story: Breaking Free from LLM Lock-In

The Challenge

You’re building an AI agent in Go. You start with OpenAI’s GPT-4 because it’s the best. But then:

  • Costs add up — Your prototype’s API bills are growing faster than your user base
  • You’re locked in — What if you want to try Anthropic’s Claude? Or use open-source Llama models?
  • Experimentation is hard — Testing different models means rewriting integration code
  • Enterprise needs — Your client wants Azure OpenAI for compliance, but your code is OpenAI-specific

You’re stuck. Switching providers means days of refactoring.

The Solution

AgenticGoKit v0.4.6 changes the game. With HuggingFace and OpenRouter support, you can now:

// Start with HuggingFace for experimentation
config.LLM.Provider = "huggingface"
config.LLM.Model = "meta-llama/Llama-3.2-1B-Instruct"

// Switch to OpenRouter to test Claude
config.LLM.Provider = "openrouter"
config.LLM.Model = "anthropic/claude-3-haiku"

// Move to production with OpenAI
config.LLM.Provider = "openai"
config.LLM.Model = "gpt-4-turbo"

Same code. Different providers. One line change.

The Impact

With 5 supported providers (OpenAI, Azure OpenAI, HuggingFace, OpenRouter, Ollama), AgenticGoKit now gives you:

Provider What You Get Why It Matters
HuggingFace (NEW) Open-source models (Llama 3.2) + Router API Free tier, experimentation, self-hosting options
OpenRouter (NEW) 50+ models (GPT, Claude, Gemini, Llama) Single API, compare models, optimize costs
OpenAI GPT-4, GPT-3.5 Industry standard, highest quality
Azure OpenAI Enterprise GPT models Compliance, SLAs, enterprise support
Ollama Local models Privacy, offline, zero API costs

Build once. Deploy anywhere. Switch anytime.

Take Action: Try It Now

Step 1: Install (30 seconds)

go get github.com/kunalkushwaha/agenticgokit@latest

Step 2: Get an API Key (2 minutes)

Pick one provider to start:

Step 3: Run Your First Agent (1 minute)

# HuggingFace - Free tier, open models
export HUGGINGFACE_API_KEY="hf_..."
cd examples/vnext/huggingface-quickstart
go run simple.go

# OR OpenRouter - 50+ models to choose from
export OPENROUTER_API_KEY="sk-or-..."
cd examples/vnext/openrouter-quickstart
go run main.go

That’s it. You’re running AI agents in Go.

What You Can Build Today

1. Cost Optimizer Agent

// Start cheap with HuggingFace Llama
config.LLM.Provider = "huggingface"
config.LLM.Model = "meta-llama/Llama-3.2-1B-Instruct"

// Upgrade to GPT-4 for complex queries
if complexQuery {
    config.LLM.Provider = "openai"
    config.LLM.Model = "gpt-4-turbo"
}

2. Multi-Model Research Assistant

// Use OpenRouter to query multiple models
models := []string{
    "openai/gpt-4-turbo",
    "anthropic/claude-3-haiku",
    "google/gemini-2.0-flash-exp:free",
}

for _, model := range models {
    config.LLM.Model = model
    // Compare responses from different models
}

3. Privacy-First Local Agent

// Start with Ollama for development
config.LLM.Provider = "ollama"
config.LLM.Model = "llama3.2"

// Switch to HuggingFace TGI for production
config.LLM.Provider = "huggingface"
config.LLM.BaseURL = "http://your-tgi-server:8080"

Learn More

Quick Links:

Join the Community

AgenticGoKit is open source and growing. We’d love your feedback:

The Bottom Line

AgenticGoKit v0.4.6 gives you freedom.

  • Start with HuggingFace for free experimentation
  • Switch to OpenRouter to test 50+ models
  • Move to OpenAI for production quality
  • Deploy to Azure OpenAI for enterprise compliance
  • Or keep it local with Ollama

One codebase. Five providers. Endless possibilities.

# Get started now
go get github.com/kunalkushwaha/agenticgokit@latest

Happy building!

Questions? Join our community discussions or check out the documentation.