Romancy

Romancy (from "Romance" - medieval chivalric tales) - Durable execution framework for Go

Lightweight durable execution framework - no separate server required

Overview

Romancy is a lightweight durable execution framework for Go that runs as a library in your application - no separate workflow server required. It provides automatic crash recovery through deterministic replay, allowing long-running workflows to survive process restarts and failures without losing progress.

Perfect for: Order processing, distributed transactions (Saga pattern), and any workflow that must survive crashes.

Romancy is a Go port of Edda (Python), providing the same core concepts and patterns in idiomatic Go.

Key Features

• ✨ Lightweight Library: Runs in your application process - no separate server infrastructure
• 🔄 Durable Execution: Deterministic replay with workflow history for automatic crash recovery
• 🎯 Workflow & Activity: Clear separation between orchestration logic and business logic
• 🔁 Saga Pattern: Automatic compensation on failure
• 🌐 Multi-worker Execution: Run workflows safely across multiple servers or containers
• 📦 Transactional Outbox: Reliable event publishing with guaranteed delivery
• ☁️ CloudEvents Support: Native support for CloudEvents protocol
• ⏱️ Event & Timer Waiting: Free up worker resources while waiting for events or timers
• 🌍 net/http Integration: Use with any Go HTTP router (chi, gorilla/mux, standard library)
• 🔒 Automatic Transactions: Activities run in transactions by default with ctx.Session() access
• 🤖 MCP Integration: Expose workflows as MCP tools for AI assistants
• 📬 Channel Messaging: Broadcast and competing message delivery between workflows
• 🔄 Recur Pattern: Erlang-style tail recursion for long-running workflows
• 📡 PostgreSQL LISTEN/NOTIFY: Real-time event delivery without polling
• 🤖 LLM Integration: Durable LLM calls via bucephalus with automatic caching for replay

Documentation

📚 Full documentation: https://i2y.github.io/romancy/

Architecture

Romancy runs as a lightweight library in your applications, with all workflow state stored in a shared database:

┌─────────────────────────────────────────────────────────────────────┐
│                          Your Go Applications                        │
├──────────────────────┬──────────────────────┬──────────────────────┤
│   order-service-1    │   order-service-2    │   order-service-3    │
│   ┌──────────────┐   │   ┌──────────────┐   │   ┌──────────────┐   │
│   │   Romancy    │   │   │   Romancy    │   │   │   Romancy    │   │
│   │   Workflow   │   │   │   Workflow   │   │   │   Workflow   │   │
│   └──────────────┘   │   └──────────────┘   │   └──────────────┘   │
└──────────┬───────────┴──────────┬───────────┴──────────┬───────────┘
           │                      │                      │
           └──────────────────────┼──────────────────────┘
                                  │
                         ┌────────▼────────┐
                         │ Shared Database │
                         │ (SQLite/PG/MySQL)│
                         └─────────────────┘

Key Points:

• Multiple workers can run simultaneously across different pods/servers
• Each workflow instance runs on only one worker at a time (automatic coordination)
• WaitEvent() and Sleep() free up worker resources while waiting
• Automatic crash recovery with stale lock cleanup and workflow auto-resume

Quick Start

package main

import (
    "context"
    "log"
    "net/http"

    "github.com/i2y/romancy"
)

// Input/Output types
type OrderInput struct {
    OrderID string  `json:"order_id"`
    Amount  float64 `json:"amount"`
}

type OrderResult struct {
    OrderID string         `json:"order_id"`
    Payment map[string]any `json:"payment"`
}

// Define an activity
var processPayment = romancy.DefineActivity("process_payment",
    func(ctx context.Context, amount float64) (map[string]any, error) {
        log.Printf("Processing payment: $%.2f", amount)
        return map[string]any{"status": "paid", "amount": amount}, nil
    },
)

// Define a workflow
var orderWorkflow = romancy.DefineWorkflow("order_workflow",
    func(ctx *romancy.WorkflowContext, input OrderInput) (OrderResult, error) {
        // Activity results are recorded in history
        result, err := processPayment.Execute(ctx, input.Amount)
        if err != nil {
            return OrderResult{}, err
        }
        return OrderResult{OrderID: input.OrderID, Payment: result}, nil
    },
)

func main() {
    ctx := context.Background()

    // Create app with SQLite storage
    app := romancy.NewApp(
        romancy.WithDatabase("workflow.db"),
        romancy.WithWorkerID("order-service"),
    )
    defer app.Shutdown(ctx)

    // Start the application
    if err := app.Start(ctx); err != nil {
        log.Fatal(err)
    }

    // Register and start workflow
    romancy.RegisterWorkflow[OrderInput, OrderResult](app, orderWorkflow)

    instanceID, err := romancy.StartWorkflow(ctx, app, orderWorkflow, OrderInput{
        OrderID: "ORD-123",
        Amount:  99.99,
    })
    if err != nil {
        log.Fatal(err)
    }
    log.Printf("Started workflow: %s", instanceID)

    // Serve CloudEvents endpoint
    http.Handle("/", app.Handler())
    log.Fatal(http.ListenAndServe(":8001", nil))
}

What happens on crash?

1. Activities already executed return cached results from history
2. Workflow resumes from the last checkpoint
3. No manual intervention required

Installation

go get github.com/i2y/romancy

Database Support

Database	Use Case	Multi-Pod Support	Production Ready
SQLite	Development, testing, single-process	⚠️ Limited	⚠️ Limited
PostgreSQL	Production, multi-process/multi-pod	✅ Yes	✅ Recommended
MySQL	Production, multi-process/multi-pod	✅ Yes	✅ Yes (8.0+)

Important: For multi-process or multi-pod deployments (K8s, Docker Compose with multiple replicas), use PostgreSQL or MySQL.

Database Migrations

Romancy automatically applies database migrations on startup. Migration files are embedded in the binary, so no external files are needed at runtime.

// Default: auto-migration enabled
app := romancy.NewApp(
    romancy.WithDatabase("postgres://user:pass@localhost/db"),
)

// Disable auto-migration (manage with dbmate CLI instead)
app := romancy.NewApp(
    romancy.WithDatabase("postgres://user:pass@localhost/db"),
    romancy.WithAutoMigrate(false),
)

Features:

• Automatic: Migrations run during app.Start() before any workflows execute
• dbmate-compatible: Uses the same schema_migrations table as dbmate CLI
• Multi-worker safe: Safe for concurrent startup across multiple pods/processes
• Embedded: Migration files are bundled into the binary via Go embed

The database schema is managed in the durax-io/schema repository, shared between Romancy (Go) and Edda (Python) for cross-framework compatibility.

Manual migration with dbmate (optional):

If you prefer to manage migrations externally:

# Install dbmate
brew install dbmate  # macOS
# or: go install github.com/amacneil/dbmate@latest

# Run migrations (SQLite)
dbmate --url "sqlite:workflow.db" --migrations-dir schema/db/migrations/sqlite up

# Run migrations (PostgreSQL)
dbmate --url "postgres://user:pass@localhost/db?sslmode=disable" --migrations-dir schema/db/migrations/postgresql up

# Run migrations (MySQL)
dbmate --url "mysql://user:pass@localhost/db" --migrations-dir schema/db/migrations/mysql up

Core Concepts

Workflows and Activities

Activity: A unit of work that performs business logic. Activity results are recorded in history.

Workflow: Orchestration logic that coordinates activities. Workflows can be replayed from history after crashes.

// Input/Output types
type EmailInput struct {
    Email   string `json:"email"`
    Message string `json:"message"`
}

type SignupResult struct {
    Status string `json:"status"`
}

// Define an activity with DefineActivity
var sendEmail = romancy.DefineActivity("send_email",
    func(ctx context.Context, input EmailInput) (bool, error) {
        // Business logic - this will be recorded
        log.Printf("Sending email to %s", input.Email)
        return true, nil
    },
)

// Define a workflow with DefineWorkflow
var userSignup = romancy.DefineWorkflow("user_signup",
    func(ctx *romancy.WorkflowContext, email string) (SignupResult, error) {
        // Orchestration logic
        _, err := sendEmail.Execute(ctx, EmailInput{Email: email, Message: "Welcome!"})
        if err != nil {
            return SignupResult{}, err
        }
        return SignupResult{Status: "completed"}, nil
    },
)

Durable Execution

Romancy ensures workflow progress is never lost through deterministic replay:

1. Activity results are recorded in a history table
2. On crash recovery, workflows resume from the last checkpoint
3. Already-executed activities return cached results from history
4. New activities continue from where the workflow left off

Key guarantees:

• Activities execute exactly once (results cached in history)
• Workflows can survive arbitrary crashes
• No manual checkpoint management required

Compensation (Saga Pattern)

When a workflow fails, Romancy can execute compensation functions for already-executed activities in reverse order:

// Define activity with inline compensation
var reserveInventory = romancy.DefineActivity("reserve_inventory",
    func(ctx context.Context, itemID string) (string, error) {
        log.Printf("Reserved %s", itemID)
        return "reservation-123", nil
    },
    romancy.WithCompensation[string, string](func(ctx context.Context, itemID string) error {
        log.Printf("Cancelled reservation for %s", itemID)
        return nil
    }),
)

Multi-worker Execution

Multiple workers can safely process workflows using database-based exclusive control:

app := romancy.NewApp(
    romancy.WithDatabase("postgres://localhost/workflows"),  // Shared database
    romancy.WithWorkerID("order-service"),
)

Features:

• Each workflow instance runs on only one worker at a time
• Automatic stale lock cleanup (5-minute timeout)
• Crashed workflows automatically resume on any available worker

Transactional Outbox

Activities are automatically transactional by default, ensuring atomicity between:

1. Activity execution (side effects)
2. History recording (replay data)
3. Event publishing (outbox table)

// Default: activity runs in a transaction
var createOrder = romancy.DefineActivity("create_order",
    func(ctx context.Context, orderID string) (map[string]any, error) {
        // If this activity fails, both the history record and any
        // outbox events will be rolled back together
        log.Printf("Creating order: %s", orderID)
        return map[string]any{"order_id": orderID, "status": "created"}, nil
    },
)

Opt-out of transactions for external API calls:

var callExternalAPI = romancy.DefineActivity("call_api",
    func(ctx context.Context, url string) (map[string]any, error) {
        // Not transactional - suitable for external calls
        resp, err := http.Get(url)
        // ...
    },
    romancy.WithTransactional(false),
)

Event & Timer Waiting

Workflows can wait for external events or timers without consuming worker resources:

// Event data type
type PaymentData struct {
    OrderID string `json:"order_id"`
    Status  string `json:"status"`
}

var paymentWorkflow = romancy.DefineWorkflow("payment_workflow",
    func(ctx *romancy.WorkflowContext, orderID string) (PaymentData, error) {
        // Wait for payment completion event (process-releasing)
        event, err := romancy.WaitEvent[PaymentData](ctx, "payment.completed")
        if err != nil {
            return PaymentData{}, err
        }
        return event.Data, nil
    },
)

// With timeout
var paymentWithTimeout = romancy.DefineWorkflow("payment_with_timeout",
    func(ctx *romancy.WorkflowContext, orderID string) (PaymentData, error) {
        event, err := romancy.WaitEvent[PaymentData](ctx, "payment.completed",
            romancy.WithEventTimeout(5*time.Minute))
        if err != nil {
            return PaymentData{}, err
        }
        return event.Data, nil
    },
)

Sleep() for time-based waiting:

var orderWithTimeout = romancy.DefineWorkflow("order_with_timeout",
    func(ctx *romancy.WorkflowContext, orderID string) (map[string]any, error) {
        // Sleep for 60 seconds
        if err := romancy.Sleep(ctx, 60*time.Second); err != nil {
            return nil, err
        }
        // Continue after sleep completes
        return checkPayment.Execute(ctx, orderID)
    },
)

Channel Messaging

Workflows can communicate with each other using durable channels with two delivery modes:

• Broadcast: All subscribers receive every message
• Competing: Each message goes to exactly one subscriber (work queue pattern)

var workerWorkflow = romancy.DefineWorkflow("worker",
    func(ctx *romancy.WorkflowContext, workerID string) (map[string]any, error) {
        // Subscribe to a channel in competing mode (work queue)
        if err := romancy.Subscribe(ctx, "tasks", romancy.ModeCompeting); err != nil {
            return nil, err
        }

        // Wait for and receive a message
        msg, err := romancy.Receive[TaskData](ctx, "tasks",
            romancy.WithReceiveTimeout(60*time.Second))
        if err != nil {
            return nil, err
        }

        log.Printf("Worker %s received task: %v", workerID, msg.Data)
        return map[string]any{"task": msg.Data}, nil
    },
)

var dispatcherWorkflow = romancy.DefineWorkflow("dispatcher",
    func(ctx *romancy.WorkflowContext, _ struct{}) (map[string]any, error) {
        // Publish a task to the channel
        if err := romancy.Publish(ctx, "tasks", TaskData{ID: "task-1"}); err != nil {
            return nil, err
        }
        return map[string]any{"dispatched": true}, nil
    },
)

// Send a direct message to a specific workflow instance
romancy.SendTo(ctx, targetInstanceID, "channel", data)

// Unsubscribe from a channel
romancy.Unsubscribe(ctx, "channel")

// Publish with metadata
romancy.Publish(ctx, "channel", data, romancy.WithMetadata(map[string]any{
    "priority": "high",
}))

Recur Pattern

Long-running workflows can use Erlang-style tail recursion to prevent unbounded history growth:

// For Recur, input and output types must be the same
type CounterState struct {
    Count      int  `json:"count"`
    Completed  bool `json:"completed"`
    FinalCount int  `json:"final_count,omitempty"`
}

var counterWorkflow = romancy.DefineWorkflow("counter",
    func(ctx *romancy.WorkflowContext, input CounterState) (CounterState, error) {
        newCount := input.Count + 1

        if newCount < 1000 {
            // Continue with new input (archives history, restarts workflow)
            return romancy.Recur(ctx, CounterState{Count: newCount})
        }

        // Complete when done
        return CounterState{Completed: true, FinalCount: newCount}, nil
    },
)

Key benefits:

• History is archived before each recur
• continued_from field tracks workflow lineage
• Prevents memory/storage issues with long-running workflows

HTTP Integration

Using with net/http

Romancy provides a Handler() method for integration with any Go HTTP router:

// Standard library
http.Handle("/", app.Handler())

// With chi router
r := chi.NewRouter()
r.Mount("/workflows", app.Handler())

// With gorilla/mux
r := mux.NewRouter()
r.PathPrefix("/workflows").Handler(app.Handler())

CloudEvents Support

Romancy accepts CloudEvents at the HTTP endpoint:

curl -X POST http://localhost:8001/ \
  -H "Content-Type: application/cloudevents+json" \
  -d '{
    "specversion": "1.0",
    "type": "order.created",
    "source": "order-service",
    "id": "event-123",
    "data": {"order_id": "ORD-123", "amount": 99.99}
  }'

HTTP Response Codes:

• 202 Accepted: Event accepted for asynchronous processing
• 400 Bad Request: CloudEvents parsing/validation error (non-retryable)
• 500 Internal Server Error: Internal error (retryable)

MCP Integration

Romancy provides Model Context Protocol (MCP) integration, allowing you to expose workflows as MCP tools for AI assistants.

Quick Start

package main

import (
	"context"
	"log"

	"github.com/i2y/romancy"
	"github.com/i2y/romancy/mcp"
)

// Define workflow input/output types with jsonschema tags
type OrderInput struct {
	OrderID    string  `json:"order_id" jsonschema:"The unique order ID"`
	CustomerID string  `json:"customer_id" jsonschema:"The customer ID"`
	Amount     float64 `json:"amount" jsonschema:"The order amount in dollars"`
}

type OrderResult struct {
	OrderID      string `json:"order_id" jsonschema:"The order ID"`
	Status       string `json:"status" jsonschema:"Final order status"`
	Confirmation string `json:"confirmation" jsonschema:"Confirmation number"`
}

// Define workflow
var orderWorkflow = romancy.DefineWorkflow("order_workflow",
	func(ctx *romancy.WorkflowContext, input OrderInput) (OrderResult, error) {
		// Workflow logic here
		return OrderResult{
			OrderID:      input.OrderID,
			Status:       "confirmed",
			Confirmation: "CONF-123",
		}, nil
	},
)

func main() {
	// Create Romancy app
	app := romancy.NewApp(romancy.WithDatabase("workflow.db"))

	// Create MCP server
	server := mcp.NewServer(app,
		mcp.WithServerName("order-service"),
		mcp.WithServerVersion("1.0.0"),
	)

	// Register workflow - auto-generates 4 MCP tools:
	// - order_workflow_start
	// - order_workflow_status
	// - order_workflow_result
	// - order_workflow_cancel
	mcp.RegisterWorkflow[OrderInput, OrderResult](server, orderWorkflow,
		mcp.WithDescription("Process customer orders"),
	)

	// Initialize and run
	ctx := context.Background()
	if err := server.Initialize(ctx); err != nil {
		log.Fatal(err)
	}
	defer server.Shutdown(ctx)

	// Run on stdio transport
	if err := server.RunStdio(ctx); err != nil {
		log.Fatal(err)
	}
}

Auto-Generated Tools

When you register a workflow, Romancy automatically generates four MCP tools:

Tool	Description
{workflow}_start	Start a new workflow instance with input parameters
{workflow}_status	Get the current status of a workflow instance
{workflow}_result	Get the result of a completed workflow
{workflow}_cancel	Cancel a running workflow instance

LLM Integration

Romancy provides durable LLM calls via the bucephalus library. All LLM calls are automatically cached as activities, so workflow replay returns cached results without re-invoking the LLM API.

Quick Start

package main

import (
	"context"
	"fmt"
	"log"

	"github.com/i2y/romancy"
	"github.com/i2y/romancy/llm"

	// Import the provider you want to use
	_ "github.com/i2y/bucephalus/anthropic"
)

type SummaryInput struct {
	Text string `json:"text"`
}

type SummaryResult struct {
	Summary string `json:"summary"`
}

var summarizeWorkflow = romancy.DefineWorkflow("summarize",
	func(ctx *romancy.WorkflowContext, input SummaryInput) (SummaryResult, error) {
		// LLM call is cached for replay
		response, err := llm.Call(ctx, input.Text,
			llm.WithSystemMessage("Summarize the following text concisely."),
		)
		if err != nil {
			return SummaryResult{}, err
		}
		return SummaryResult{Summary: response.Text}, nil
	},
)

func main() {
	ctx := context.Background()

	// Create app
	app := romancy.NewApp(romancy.WithDatabase("workflow.db"))

	// Set LLM defaults for the app
	llm.SetAppDefaults(app,
		llm.WithProvider("anthropic"),
		llm.WithModel("claude-sonnet-4-5-20250929"),
		llm.WithMaxTokens(1024),
	)

	romancy.RegisterWorkflow(app, summarizeWorkflow)

	if err := app.Start(ctx); err != nil {
		log.Fatal(err)
	}
	defer app.Shutdown(ctx)

	instanceID, _ := romancy.StartWorkflow(ctx, app, summarizeWorkflow, SummaryInput{
		Text: "Long article text here...",
	})
	fmt.Println("Started:", instanceID)
}

Features

Feature	Description
llm.Call()	Ad-hoc LLM call with automatic caching
llm.CallParse[T]()	Parse response into a struct
llm.CallMessages()	Multi-turn conversation
llm.DefineDurableCall()	Reusable LLM definition
llm.DurableAgent[T]	Stateful multi-turn agent
llm.SetAppDefaults()	App-level default configuration

Supported Providers

Via bucephalus, Romancy supports:

• Anthropic (Claude models)
• OpenAI (GPT models)
• Google Gemini

Development

Running Tests

go test ./...

Project Structure

romancy/
├── app.go           # Main App struct and HTTP handling
├── workflow.go      # Workflow definition and execution
├── activity.go      # Activity definition with transactions
├── context.go       # WorkflowContext for workflow execution
├── events.go        # Event and timer waiting
├── compensation.go  # Saga pattern compensation
├── hooks/           # Observability hooks
├── outbox/          # Transactional outbox
├── mcp/             # MCP (Model Context Protocol) integration
├── llm/             # LLM integration (via bucephalus)
└── internal/
    ├── storage/     # Database abstraction
    └── replay/      # Deterministic replay engine

License

This project is licensed under the MIT License - see the LICENSE file for details.

Related Projects

• Edda: The original Python implementation