github.com/scottdavis/dsgo

DSGo

This is mostly for my personal use, so support is not guaranteed.

A Go implementation of the Declarative Self-improving programming (DSPy) pattern.

Acknowledgments

I would like to express my sincere gratitude to Xiao Constantine (@XiaoConstantine), the original author of dspy-go. This project builds upon his excellent foundation and pioneering work in bringing the DSPy pattern to the Go ecosystem.

Recent Improvements

This fork includes several enhancements to the original project:

• Advanced LLM Configuration: Added more flexible ways to configure Ollama with custom hosts via multiple format options
• Ability to inject llm instances: into functions so you cna use different llms for different parts/steps of your program
• OpenRouter Integration: Full support for OpenRouter as an LLM provider, enabling access to hundreds of AI models through a single interface
• Interface for Memory: Added an interface for memory, allowing you to use different memory backends, including Redis

DSGo is a Go implementation of DSPy, bringing systematic prompt engineering and automated reasoning capabilities to Go applications. It provides a flexible framework for building reliable and effective Language Model (LLM) applications through composable modules and workflows.

Installation

go get github.com/scottdavis/dsgo

Quick Start

Here's a simple example to get you started with DSGo:

package main

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

	"github.com/scottdavis/dsgo/pkg/core"
	"github.com/scottdavis/dsgo/pkg/llms"
	"github.com/scottdavis/dsgo/pkg/modules"
	"github.com/scottdavis/dsgo/pkg/logging"
)

func main() {
	// Setup DSP-GO logging
	output := logging.NewConsoleOutput(true, logging.WithColor(true))
	logger := logging.NewLogger(logging.Config{
		Severity: logging.INFO,
		Outputs:  []logging.Output{output},
	})
	logging.SetLogger(logger)

	// Create basic context
	basicCtx := context.Background()

	// Initialize LLM (example with Ollama)
	ollamaConfig := llms.NewOllamaConfig("http://localhost:11434", "llama3")
	llm, err := llms.NewLLM("", ollamaConfig)
	if err != nil {
		log.Fatalf("Failed to initialize LLM: %v", err)
		os.Exit(1)
	}

	// Create DSP-GO context with execution state
	ctx := core.WithExecutionState(basicCtx)

	// Create a signature for question answering
	signature := core.NewSignature(
		[]core.InputField{{Field: core.Field{Name: "question"}}},
		[]core.OutputField{{Field: core.Field{Name: "answer"}}},
	)

	// Create a DSPYConfig with the LLM
	dspyConfig := core.NewDSPYConfig().WithDefaultLLM(llm)

	// Create a ChainOfThought module with the config
	cot := modules.NewChainOfThought(signature, dspyConfig)

	// Create a program
	program := core.NewProgram(
		map[string]core.Module{"cot": cot},
		func(ctx context.Context, inputs map[string]interface{}) (map[string]interface{}, error) {
			return cot.Process(ctx, inputs)
		},
		dspyConfig,
	)

	// Execute the program
	result, err := program.Execute(ctx, map[string]interface{}{
		"question": "What is the capital of France?",
	})
	if err != nil {
		log.Fatalf("Error executing program: %v", err)
	}

	fmt.Printf("Answer: %s\n", result["answer"])
}

Core Concepts

Signatures

Signatures define the input and output fields for modules. They help in creating type-safe and well-defined interfaces for your AI components.

Modules

Modules are the building blocks of DSGo programs. They encapsulate specific functionalities and can be composed to create complex pipelines. Some key modules include:

• Predict: Basic prediction module
• ChainOfThought: Implements chain-of-thought reasoning
• ReAct: Implements the ReAct (Reasoning and Acting) paradigm

Optimizers

Optimizers help improve the performance of your DSGo programs by automatically tuning prompts and module parameters. Including:

• BootstrapFewShot: Automatic few-shot example selection
• MIPRO: Multi-step interactive prompt optimization
• Copro: Collaborative prompt optimization

Agents

Use dspy's core concepts as building blocks, impl Building Effective Agents

• Chain Workflow: Sequential execution of steps
• Parallel Workflow: Concurrent execution with controlled parallelism
• Router Workflow: Dynamic routing based on classification
• Orchestrator: Flexible task decomposition and execution

See agent examples

// Chain
workflow := workflows.NewChainWorkflow(store)
workflow.AddStep(&workflows.Step{
    ID: "step1",
    Module: modules.NewPredict(signature1),
})
workflow.AddStep(&workflows.Step{
    ID: "step2", 
    Module: modules.NewPredict(signature2),
})

Each workflow step can be configured with:

• Retry logic with exponential backoff
• Conditional execution based on workflow state
• Custom error handling

step := &workflows.Step{
    ID: "retry_example",
    Module: myModule,
    RetryConfig: &workflows.RetryConfig{
        MaxAttempts: 3,
        BackoffMultiplier: 2.0,
    },
    Condition: func(state map[string]interface{}) bool {
        return someCondition(state)
    },
}

Advanced Features

Tracing and Logging

// Enable detailed tracing
ctx = core.WithExecutionState(context.Background())

// Configure logging
logger := logging.NewLogger(logging.Config{
    Severity: logging.DEBUG,
    Outputs:  []logging.Output{logging.NewConsoleOutput(true)},
})
logging.SetLogger(logger)

Custom Tools

You can extend ReAct modules with custom tools:

func (t *CustomTool) CanHandle(action string) bool {
    return strings.HasPrefix(action, "custom_")
}

func (t *CustomTool) Execute(ctx context.Context, action string) (string, error) {
    // Implement tool logic
    return "Tool result", nil
}

Working with Different LLM Providers

// Using Anthropic Claude
anthropicConfig := llms.NewAnthropicConfig(llms.AnthropicModelSonnet)
llm, _ := llms.NewLLM("api-key", anthropicConfig)

// Using Ollama (various options)
// Option 1: Using NewLLM with OllamaConfig
ollamaConfig := llms.NewOllamaConfig("http://localhost:11434", "llama2")
llm, _ := llms.NewLLM("", ollamaConfig)

// Option 2: Using OpenRouter
openRouterConfig := llms.NewOpenRouterConfig(llms.OpenRouterModelNeevaAI)
llm, _ := llms.NewLLM("api-key", openRouterConfig)

// Option 3: Using LlamaCPP
llm, _ := llms.NewLlamacppLLM("http://localhost:8080")

// Option 4: Using string format with custom host
llm, _ := llms.NewLLM("", "ollama:example.com:11434:llama2")
// or with protocol
llm, _ := llms.NewLLM("", "ollama:http://example.com:11434:llama2")

// Create DSPYConfig with the LLM
dspyConfig := core.NewDSPYConfig().WithDefaultLLM(llm)

Examples

Check the examples directory for complete implementations:

• examples/agents: Demonstrates different agent patterns
• examples/hotpotqa: Question-answering implementation
• examples/gsm8k: Math problem solving

License

DSGo is released under the MIT License. See the LICENSE file for details.

Redis Memory Store Example

This example demonstrates how to use Redis as a memory store for DSP-GO applications. The example shows a practical workflow where data is downloaded from an external source, stored in Redis, and then processed in subsequent steps.

Features Demonstrated

• Redis Connection: Setting up and connecting to a Redis server
• Data Storage: Storing data in Redis between processing steps
• Data Retrieval: Retrieving and processing the stored data
• TTL (Time-to-Live): Using TTL functionality for automatic expiration of data
• Cleanup: Properly cleaning up Redis resources

Running the Example

Prerequisites

• Redis server running locally or accessible remotely
• Go 1.18 or later

Start Redis

If you don't have Redis running, you can quickly start it with Docker:

docker run -d --name redis-example -p 6379:6379 redis:latest

Run the Example

go run main.go

By default, it connects to Redis at localhost:6379. You can specify a different Redis server:

go run main.go --redis-addr=your-redis-server:6379 --redis-password=yourpassword

You can also specify a different URL to download:

go run main.go --url=https://raw.githubusercontent.com/your-repo/your-file.txt

How It Works

The example performs these steps:

• Connect to Redis: Establishes a connection to the Redis server
• Download Content: Fetches content from a specified URL
• Store in Redis: Stores the downloaded content in Redis
• Process Content: Retrieves the content from Redis and processes it (counts lines and words)
• Demonstrate TTL: Shows how to use TTL functionality to automatically expire data
• Clean Up: Properly cleans up Redis resources

Code Structure

• main.go: The main application code demonstrating Redis usage
• main_test.go: Tests for the Redis functionality

Redis Memory Options in DSP-GO

DSP-GO provides several memory store implementations including:

• Redis: Used in this example, provides persistent storage with TTL support
• In-Memory: Simple in-memory storage
• SQLite: Local disk-based storage

Redis is particularly useful for:

• High-performance persistence between application restarts
• Distributed applications needing shared memory
• Scenarios requiring automatic data expiration