A lightweight Go framework for building ReAct pattern agents — inspired by LangChain. It lets you combine LLMs, tools, and structured prompts to create agentic software.

✅ Implements the ReAct loop (Reason + Act + Observe) ✅ Pluggable LLM interface (llm.LLM) ✅ Extensible tools system (tools.Tool) ✅ Built-in prompt engineering system (prompt package) ✅ Ready-to-use tools: Calculator, Search ✅ Simple developer API for fast prototyping

Install in your Go project:

go get github.com/Rustixir/aylar

Example:

package examples

import (
	"aylar/agent"
	"aylar/llm"
	"aylar/tool"
	"fmt"
	"os"
)

func main() {
	openaiLLM := &llm.OpenAI{ApiKey: os.Getenv("OPENAI_API_KEY")}

	tools := []tool.Interface{
		new(tool.Calculator),
		new(tool.Search),
	}

	agent := agent.NewAgent(openaiLLM, tools)

	question := "What is 5 * 12 ? And who is the president of the USA?"
	answer, err := agent.Run(question)
	if err != nil {
		fmt.Println("Error:", err)
		return
	}

	fmt.Println("Agent Final Answer:", answer)
}

The prompt package helps you design prompts intentionally, not just pass raw text.

✅ PromptTemplate → structured templates with variables

✅ FewShotPrompt → include example Q&A pairs to guide the LLM

✅ ReActPromptBuilder → generate well-formed ReAct prompts that explain available tools and current context

This gives you control and consistency over how the agent interacts with the LLM, helping it reason and choose actions effectively.

type JokeTool struct{}
func (j *JokeTool) Name() string { return "Joke" }
func (j *JokeTool) Description() string { return "Tells a joke" }
func (j *JokeTool) Run(input string) (string, error) {
    return "Why did the chicken cross the road? To get to the other side!", nil
}

go test ./...

MIT License

If you want, I can generate the full README.md file for you, ready to drop into your repo — should I do that?