lingllm

LingLLM

A universal Go library for building LLM applications with support for multiple providers, tools, chains, and streaming.

Features

• Multi-Provider Support: Unified interface for OpenAI, Anthropic, and other LLM providers
• Tool/Function Calling: Built-in support for tool execution and management
• Streaming: Full support for streaming responses with event-based processing
• Chain Architecture: Composable chain-based processing pipeline for complex workflows
• Tool Chains: Automatic tool calling and result collection with configurable rounds
• Metrics: Built-in metrics collection for monitoring API calls and performance
• Type-Safe: Strongly typed Go interfaces and implementations
• Embeddings: Multi-provider text embedding (OpenAI, Ollama, Nvidia, DashScope, Local)
• Full-Text Search: Bleve-powered search with facets, highlighting, and suggestions
• Document Retrieval: Multi-strategy retrieval (vector, keyword, hybrid) with reranking
• Chunking: Intelligent document chunking with multiple strategies
• Knowledge Base: Integrated knowledge management with vector databases (Qdrant, Milvus)

Installation

go get github.com/LingByte/lingllm

Quick Start

Basic Chat

package main

import (
	"context"
	"fmt"
	"github.com/LingByte/lingllm/protocol"
)

func main() {
	// Create a chat request
	req := protocol.NewChatRequest(
		"gpt-4",
		protocol.UserMessage("What is the capital of France?"),
	)

	// Call your model implementation
	// resp, err := model.Chat(context.Background(), *req)
	// if err != nil {
	//     panic(err)
	// }
	// fmt.Println(resp.FirstContent())
}

Tool Calling

package main

import (
	"context"
	"encoding/json"
	"fmt"
	"github.com/LingByte/lingllm/protocol"
	"github.com/LingByte/lingllm/tools"
)

func main() {
	// Create a tool executor
	executor := tools.NewSimpleToolExecutor()

	// Register a tool
	weatherTool := tools.WeatherTool()
	executor.RegisterTool(weatherTool, func(args json.RawMessage) (string, error) {
		// Implement weather lookup
		return "Sunny, 72°F", nil
	})

	// Create a tool chain
	toolChain := tools.NewToolChain(model, executor)
	toolChain.WithMaxRounds(5)

	// Execute with tools
	req := protocol.NewChatRequest(
		"gpt-4",
		protocol.UserMessage("What's the weather in San Francisco?"),
	)
	resp, err := toolChain.ExecuteWithTools(context.Background(), *req)
	if err != nil {
		panic(err)
	}
	fmt.Println(resp.FirstContent())
}

Streaming

package main

import (
	"context"
	"fmt"
	"io"
	"github.com/LingByte/lingllm/protocol"
)

func main() {
	req := protocol.NewChatRequest(
		"gpt-4",
		protocol.UserMessage("Write a poem about Go programming"),
	)

	stream, err := model.StreamChat(context.Background(), *req)
	if err != nil {
		panic(err)
	}
	defer stream.Close()

	for {
		chunk, err := stream.Recv()
		if err == io.EOF {
			break
		}
		if err != nil {
			panic(err)
		}
		fmt.Print(chunk.Delta)
	}
}

Chains

package main

import (
	"context"
	"github.com/LingByte/lingllm/chain"
	"github.com/LingByte/lingllm/protocol"
)

func main() {
	// Build a chain with multiple models/processors
	c := chain.NewBuilder("my-chain").
		AddModel("model1", model1).
		AddProcessor("processor1", func(ctx context.Context, resp *protocol.ChatResponse) (*protocol.ChatResponse, error) {
			// Process response
			return resp, nil
		}).
		AddModel("model2", model2).
		Build()

	req := protocol.NewChatRequest(
		"gpt-4",
		protocol.UserMessage("Hello"),
	)

	resp, err := c.Invoke(context.Background(), *req)
	if err != nil {
		panic(err)
	}
	println(resp.FirstContent())
}

Embeddings

package main

import (
	"context"
	"github.com/LingByte/lingllm/embedder"
)

func main() {
	// Create embedder with OpenAI
	cfg := &embedder.Config{
		Provider: "openai",
		Model:    "text-embedding-3-small",
		APIKey:   os.Getenv("OPENAI_API_KEY"),
	}
	
	emb, err := embedder.Create(context.Background(), cfg)
	if err != nil {
		panic(err)
	}
	defer emb.Close()

	// Single embedding
	vec, err := emb.EmbedSingle(context.Background(), "Hello world")
	if err != nil {
		panic(err)
	}
	fmt.Printf("Vector dimension: %d\n", len(vec))

	// Batch embedding
	vecs, err := emb.Embed(context.Background(), []string{
		"Hello world",
		"Goodbye world",
	})
	if err != nil {
		panic(err)
	}
	fmt.Printf("Embedded %d texts\n", len(vecs))
}

Full-Text Search

package main

import (
	"context"
	"github.com/LingByte/lingllm/search"
)

func main() {
	// Create search engine
	cfg := search.Config{
		IndexPath:           "./search_index",
		DefaultAnalyzer:     "standard",
		DefaultSearchFields: []string{"title", "body"},
	}
	
	m := search.BuildIndexMapping("standard")
	engine, err := search.New(cfg, m)
	if err != nil {
		panic(err)
	}
	defer engine.Close()

	// Index documents
	docs := []search.Doc{
		{
			ID:   "1",
			Type: "article",
			Fields: map[string]interface{}{
				"title": "Go Programming",
				"body":  "Go is a fast and efficient language",
			},
		},
	}
	engine.IndexBatch(context.Background(), docs)

	// Search
	result, err := engine.Search(context.Background(), search.SearchRequest{
		Keyword: "Go",
		Size:    10,
	})
	if err != nil {
		panic(err)
	}
	
	fmt.Printf("Found %d results\n", result.Total)
	for _, hit := range result.Hits {
		fmt.Printf("- %s (score: %.2f)\n", hit.Fields["title"], hit.Score)
	}
}

Document Retrieval

package main

import (
	"context"
	"github.com/LingByte/lingllm/retrieve"
)

func main() {
	// Create hybrid retriever
	retriever, err := retrieve.New(retrieve.Config{
		Strategy:     retrieve.StrategyHybrid,
		Vector:       vectorStore,
		Search:       searchEngine,
		TopK:         10,
		VectorWeight: 0.65,
	})
	if err != nil {
		panic(err)
	}

	// Retrieve documents
	docs, err := retriever.Retrieve(context.Background(), "machine learning", 10)
	if err != nil {
		panic(err)
	}
	
	for i, doc := range docs {
		fmt.Printf("%d. %s (score: %.2f)\n", i+1, doc.Content, doc.Score)
	}
}

Knowledge Base

package main

import (
	"context"
	"github.com/LingByte/lingllm/knowledge"
	"github.com/LingByte/lingllm/embedder"
	"github.com/LingByte/lingllm/search"
)

func main() {
	// Create embedder
	emb, _ := embedder.Create(context.Background(), &embedder.Config{
		Provider: "openai",
		Model:    "text-embedding-3-small",
		APIKey:   os.Getenv("OPENAI_API_KEY"),
	})

	// Create search engine
	searchCfg := search.Config{
		IndexPath:           "./search_index",
		DefaultSearchFields: []string{"title", "content"},
	}
	m := search.BuildIndexMapping("standard")
	searcher, _ := search.New(searchCfg, m)

	// Create vector database handler
	handler, _ := knowledge.NewKnowledgeHandler(knowledge.HandlerFactoryParams{
		Provider: knowledge.ProviderQdrant,
		QdrantConfig: &knowledge.QdrantConfig{
			BaseURL: "http://localhost:6333",
			APIKey:  "your-api-key",
		},
	})

	// Create knowledge base
	kb, _ := knowledge.NewKnowledgeBase(knowledge.KnowledgeBaseConfig{
		Handler:  handler,
		Embedder: emb,
		Searcher: searcher,
	})
	defer kb.Close()

	// Add document
	kb.AddDocument(context.Background(), "doc1", "Title", "Content...", nil)

	// Query
	results, _ := kb.Query(context.Background(), "search query", 10)
	for _, result := range results {
		fmt.Printf("%s (score: %.2f)\n", result.Record.Title, result.Score)
	}
}

Project Structure

lingllm/
├── protocol/          # Core protocol definitions and interfaces
│   ├── types.go       # ChatRequest, ChatResponse, Message, Tool definitions
│   ├── factory.go     # Provider factory for creating clients
│   ├── stream.go      # Streaming utilities and transformers
│   └── ...
├── tools/             # Tool execution and management
│   ├── tools.go       # Tool definitions, executors, and chains
│   └── ...
├── chain/             # Chain-based processing pipeline
│   ├── chain.go       # Chain composition and execution
│   └── ...
├── metrics/           # Metrics collection
│   └── metrics.go     # Call metrics and monitoring
├── embedder/          # Text embedding providers
│   ├── types.go       # Core types and interfaces
│   ├── local.go       # Local MD5-based embedder
│   ├── openai.go      # OpenAI embeddings
│   ├── ollama.go      # Ollama local embeddings
│   ├── nvidia.go      # Nvidia enterprise embeddings
│   ├── dashscope.go   # Alibaba DashScope embeddings
│   ├── factory.go     # Embedder factory pattern
│   └── *_test.go      # Comprehensive tests (80%+ coverage)
├── search/            # Full-text search engine
│   ├── types.go       # Search types and interfaces
│   ├── engine.go      # Bleve-based search engine
│   ├── mapping.go     # Index mapping configuration
│   ├── query_builder.go # Query construction
│   └── *_test.go      # Tests (96.1% coverage)
├── retrieve/          # Multi-strategy document retrieval
│   ├── types.go       # Retrieval types
│   ├── config.go      # Configuration and factory
│   ├── retriever.go   # Retrieval logic
│   └── *_test.go      # Tests (80.6% coverage)
├── chunk/             # Document chunking strategies
│   └── ...
├── knowledge/         # Knowledge base management
│   ├── types.go       # Core types and interfaces
│   ├── integration.go # KnowledgeBase integration
│   ├── qdrant.go      # Qdrant vector database handler
│   ├── milvus.go      # Milvus vector database handler
│   ├── embedding.go   # Embedding client (Nvidia)
│   ├── doc_type_detector.go # Document type detection
│   ├── README.md      # Knowledge base documentation
│   └── *_test.go      # Tests (40+ tests)
├── utils/             # Shared utilities
│   ├── clean.go       # Text cleaning utilities
│   └── *_test.go      # Tests
├── shared/            # Shared utilities
├── examples/          # Example implementations
│   ├── embedder-demo/ # Multi-provider embedding demo
│   ├── search-demo/   # Full-text search demo
│   └── ...
└── go.mod

Core Concepts

Protocol

The protocol package defines the core interfaces and types:

• ChatModel: Interface for language models
• ChatRequest: Unified request format
• ChatResponse: Normalized response format
• ChatStream: Streaming interface
• Tool: Tool/function definitions

Tools

The tools package provides:

• ToolExecutor: Interface for executing tools
• SimpleToolExecutor: Basic implementation using function maps
• ToolChain: Automatic tool calling with conversation management
• ToolCallParser: Parse tool calls from model responses (ReAct, JSON formats)

Chains

The chain package enables:

• Chain: Sequential composition of nodes
• Node: Composable units (models, processors, stream processors)
• Builder: Fluent API for building chains
• ProcessingStream: Stream transformation pipeline

Metrics

The metrics package tracks:

• API call latency
• Token usage
• Error rates
• Provider-specific metrics

Embeddings

The embedder package provides multi-provider text embeddings:

• OpenAI: High-quality embeddings (1536 dimensions)
• Ollama: Local deployment support (384 dimensions)
• Nvidia: Enterprise-grade embeddings (1024+ dimensions)
• DashScope: Alibaba's native API (64-2048 dimensions)
• Local: MD5-based deterministic embeddings (384 dimensions)

Features:

• Unified interface across providers
• Batch processing support
• Configurable dimensions
• Vector normalization
• Factory pattern for easy provider switching

Search

The search package provides full-text search powered by Bleve:

• Full-Text Search: Keyword and phrase matching
• Advanced Queries: Match, phrase, prefix, wildcard, regex, fuzzy
• Faceted Search: Category aggregation and counting
• Highlighting: Query term highlighting with HTML formatting
• Suggestions: Autocomplete and search suggestions
• Pagination: Offset-based result pagination
• Sorting: Multi-field sorting support

Retrieval

The retrieve package implements multi-strategy document retrieval:

• Vector Strategy: Dense vector similarity search
• Keyword Strategy: Full-text search
• Hybrid Strategy: Combines vector and keyword with configurable weights
• Reranking: Optional document re-scoring
• Min Score Filtering: Quality control for results

Knowledge Base

The knowledge package provides integrated knowledge management:

• Multi-Backend Support: Qdrant and Milvus vector databases
• Intelligent Chunking: Automatic document type detection and optimal chunking
• Semantic Search: Vector-based similarity with embeddings
• Full-Text Search: Keyword-based search integration
• Hybrid Retrieval: Combine vector and keyword search
• Document Management: Add, query, delete, and list documents
• Metadata Support: Flexible metadata storage and filtering
• Document Type Detection: Structured, Table/KV, Unstructured

Features:

• Automatic document chunking based on type
• Embedding generation for semantic search
• Integration with search engines
• Multi-tenancy with namespaces
• Health checks and monitoring

Supported Providers

The library provides a unified interface for:

• OpenAI (GPT-4, GPT-3.5, etc.)
• Anthropic (Claude)
• Local models (via compatible APIs)
• Custom implementations

Message Types

// Create messages
msg := protocol.UserMessage("Hello")
msg := protocol.SystemMessage("You are a helpful assistant")
msg := protocol.AssistantMessage("Hi there!")
msg := protocol.ToolMessage("Result", "tool_call_id")

// Build requests
req := protocol.NewChatRequest("gpt-4", msg1, msg2, msg3)
req.WithMaxTokens(1000).
    WithTemperature(0.7).
    WithTopP(0.9).
    WithStop("END")

Error Handling

resp, err := model.Chat(ctx, req)
if err != nil {
	// Handle error
	fmt.Printf("Error: %v\n", err)
}

Configuration

Configure clients using provider-specific implementations:

// Example with environment variables
apiKey := os.Getenv("OPENAI_API_KEY")
// Create your provider-specific client

Testing

Run tests:

go test ./...

Run tests with coverage:

go test -cover ./...

Contributing

Contributions are welcome! Please:

1. Fork the repository
2. Create a feature branch
3. Add tests for new functionality
4. Submit a pull request

License

MIT License - see LICENSE file for details

Modules Status

✅ Completed

• Embedder Module (80%+ coverage)

  • 5 providers: OpenAI, Ollama, Nvidia, DashScope, Local
  • Batch processing, vector normalization
  • Comprehensive tests and demo

• Search Module (96.1% coverage)

  • Bleve-powered full-text search
  • Advanced query types
  • Facets, highlighting, suggestions
  • 60+ tests

• Retrieve Module (80.6% coverage)

  • 3 strategies: Vector, Keyword, Hybrid
  • Reranking support
  • 26 tests

• Knowledge Base Module (40+ tests)

  • Multi-backend support (Qdrant, Milvus)
  • Intelligent document chunking
  • Semantic and keyword search
  • Document type detection
  • Metadata support
  • Health checks

• Chunk Module

  • Multiple chunking strategies
  • Configurable chunk sizes
  • Overlap support

📋 Roadmap

• Official OpenAI provider implementation
• Official Anthropic provider implementation
• MCP (Model Context Protocol) integration
• Caching layer for responses and embeddings
• Advanced prompt engineering utilities
• Evaluation framework
• More tool examples and templates
• Vector database integration (Qdrant, Milvus)
• Knowledge base management utilities
• RAG (Retrieval-Augmented Generation) pipeline

Support

For issues, questions, or suggestions, please open an issue on GitHub.