Provider-Agnostic Go AI Library

Build AI-powered Go applications with swappable providers, modular interfaces, and production-ready tooling.

import (
    "github.com/bachtiarpanjaitan/ihandai-go"
    _ "github.com/bachtiarpanjaitan/ihandai-go/plugins/ollama"
)

ai, _ := ihandai.New(
    ihandai.WithLLM("ollama", llm.WithModel("llama3")),
    ihandai.WithEmbedding("ollama", embedding.WithModel("nomic-embed-text")),
    ihandai.WithMemory(memory.NewInMemoryStore()),
)
defer ai.Close()

resp, _ := ai.Ask(ctx, "What is RAG?", ihandai.WithTopK(5))

Installation

go get github.com/bachtiarpanjaitan/ihandai-go

Features

Feature	Description
🔍 RAG Pipeline	Load → Split → Embed → Search → Rerank → Chat
🎯 Retrieval Strategies	TopK, MMR (diversity), MultiQuery (expansion)
🤖 Agents	ReAct agent loop with tool calling, retry, reflection
🧠 Memory	Multi-turn conversations with token-aware window
⚙️ Workflows	DAG-based parallel execution, conditional branching
🔌 MCP	Model Context Protocol client + filesystem server
📡 Streaming	Real-time token streaming via channels
🛡️ Production	Rate limiter, circuit breaker, tracing

Quick Links

• 📖 Usage Guide — complete API reference with examples
• 🏗️ Architecture — high-level design
• 📦 Package Structure — package layout
• 🧩 Interfaces — core interface definitions
• 🔄 Pipeline — data flow design
• 🔌 Plugin System — provider registration
• 🗺️ Roadmap — development phases

Supported Providers

LLM & Embedding

Provider	Type	Interface
Ollama	Local	ChatCompleter, Embedder
OpenAI	Cloud	ChatCompleter, Embedder (planned)
Anthropic	Cloud	ChatCompleter (planned)
Google Gemini	Cloud	ChatCompleter (planned)

Vector Stores

Provider	Type
Qdrant	Self-hosted / Cloud (planned)
pgvector	PostgreSQL extension (planned)
Milvus	Self-hosted / Cloud (planned)

Core Principles

• Interface-first — small interfaces (1-3 methods), idiomatic Go
• Provider-agnostic — database/sql-style registry, swap without code changes
• Context-aware — context.Context as first parameter in all I/O
• Concurrency-safe — sync.RWMutex, immutable-after-creation
• Structured errors — typed errors: RateLimitError, AuthError, TimeoutError
• Production-ready — tracing, rate limiting, circuit breaker

Quick Example: RAG Chatbot

package main

import (
    "bufio"
    "context"
    "fmt"
    "log"
    "os"

    "github.com/bachtiarpanjaitan/ihandai-go"
    "github.com/bachtiarpanjaitan/ihandai-go/pkg/llm"
    "github.com/bachtiarpanjaitan/ihandai-go/pkg/embedding"
    "github.com/bachtiarpanjaitan/ihandai-go/pkg/memory"
    _ "github.com/bachtiarpanjaitan/ihandai-go/plugins/ollama"
)

func main() {
    ai, _ := ihandai.New(
        ihandai.WithLLM("ollama", llm.WithModel("llama3")),
        ihandai.WithEmbedding("ollama", embedding.WithModel("nomic-embed-text")),
        ihandai.WithVectorStore("mock"),
        ihandai.WithMemory(memory.NewInMemoryStore()),
    )
    defer ai.Close()

    scanner := bufio.NewScanner(os.Stdin)
    session := "default"
    for {
        fmt.Print("> ")
        scanner.Scan()
        resp, _ := ai.AskConversation(context.Background(), session, scanner.Text())
        fmt.Println(resp.Content)
    }
}

License & Contributing

• 📄 Contributing Guide
• 📐 Architecture Decisions
• 🧪 18 packages, 100% test pass, race detector clean