Configuration

Strategos integrates with proven .NET infrastructure. This reference covers setup and configuration for each integration.

Overview

Integration	Purpose	Required
Wolverine	Saga orchestration, message routing, transactional outbox	Yes
Marten	Event sourcing, projections, time-travel queries	Yes
PostgreSQL	Persistence for both Wolverine and Marten	Yes
Microsoft.Extensions.AI	LLM integration via IChatClient	For AI workflows

---

Wolverine

Wolverine provides saga orchestration for workflow execution.

How Workflows Become Sagas

Every workflow definition is compiled into a Wolverine saga at build time:

Workflow Definition -> Source Generator -> Wolverine Saga
     (DSL)              (compile-time)      (runtime)

The source generator produces:

• A saga class with [SagaIdentity] for workflow instance tracking
• Command handlers for each step transition
• Automatic message cascading between steps

Generated Message Types

// Generated command for a step
public record ExecuteProcessPaymentCommand(
    [property: SagaIdentity] Guid WorkflowId);

// Handler in the saga
public async Task<ExecuteFulfillOrderCommand> Handle(
    ExecuteProcessPaymentCommand command,
    ProcessPayment step,
    CancellationToken ct)
{
    var result = await step.ExecuteAsync(State, ct);
    State = StateReducer.Reduce(State, result.StateUpdate);
    return new ExecuteFulfillOrderCommand(WorkflowId);
}

Transactional Outbox

Wolverine's transactional outbox ensures state updates and outgoing messages are committed atomically. If a process crashes after updating state but before sending the next message, the message is recovered from the outbox on restart.

Retry Configuration

services.AddWolverine(opts =>
{
    opts.Handlers.OnException<HttpRequestException>()
        .RetryWithCooldown(100.Milliseconds(), 500.Milliseconds(), 1.Seconds());
});

---

Marten

Marten provides event sourcing and projections.

Event Types

Event	When Emitted
WorkflowStarted	Workflow instance created
PhaseChanged	Transition between phases
StepCompleted	Step finished executing
BranchTaken	Routing decision made
ApprovalRequested	Workflow paused for human input
ApprovalReceived	Human input received
WorkflowCompleted	Workflow reached terminal state

Appending Events

session.Events.Append(
    workflowId,
    new PhaseChanged(workflowId, fromPhase, toPhase, timestamp));

Time-Travel Queries

Reconstruct state at any point in history:

// Get state at a specific version
var historicalState = await session.Events
    .AggregateStreamAsync<WorkflowState>(
        workflowId,
        version: 5);

// Get state at a specific timestamp
var historicalState = await session.Events
    .AggregateStreamAsync<WorkflowState>(
        workflowId,
        timestamp: specificTime);

Projections

Query workflows by phase or other criteria:

var awaitingApproval = await session
    .Query<WorkflowReadModel>()
    .Where(w => w.CurrentPhase == Phase.AwaitingApproval)
    .ToListAsync();

---

PostgreSQL

Both Wolverine and Marten use PostgreSQL as their persistence layer.

Database Tables

Table/Schema	Content
wolverine_*	Saga state, outbox messages, dead letters
mt_events	Event streams (append-only)
mt_doc_*	Document projections (read models)
mt_streams	Stream metadata and version tracking

Connection Configuration

var connectionString = builder.Configuration.GetConnectionString("Postgres");

services.AddMarten(opts =>
{
    opts.Connection(connectionString);
});

services.AddWolverineWithMarten(opts =>
{
    opts.PersistenceConnection(connectionString);
});

---

Microsoft.Extensions.AI

Integration for LLM-powered workflow steps.

IChatClient Integration

public class AnalyzeStep : IAgentStep<DocumentState>
{
    private readonly IChatClient _chatClient;

    public AnalyzeStep(IChatClient chatClient)
    {
        _chatClient = chatClient;
    }

    public async Task<StepResult<DocumentState>> ExecuteAsync(
        DocumentState state,
        AgentStepContext context,
        CancellationToken ct)
    {
        var response = await _chatClient.GetResponseAsync(
            $"Analyze: {state.Content}",
            ct);

        return state.With(s => s.Analysis, response).AsResult();
    }
}

Supported Providers

Provider	Package
OpenAI	Microsoft.Extensions.AI.OpenAI
Azure OpenAI	Microsoft.Extensions.AI.AzureOpenAI
Ollama	OllamaChatClient
Custom	Implement IChatClient

Streaming Responses

public class StreamingStep : IAgentStep<ChatState>
{
    public async Task<StepResult<ChatState>> ExecuteAsync(
        ChatState state,
        AgentStepContext context,
        CancellationToken ct)
    {
        await foreach (var chunk in _chatClient.GetStreamingResponseAsync(...))
        {
            context.StreamingCallback?.OnToken(chunk.Text);
        }
        // ...
    }
}

---

Complete Setup Example

Full configuration with all integrations:

var builder = WebApplication.CreateBuilder(args);

// PostgreSQL connection
var connectionString = builder.Configuration.GetConnectionString("Postgres");

// Marten (event sourcing)
builder.Services.AddMarten(opts =>
{
    opts.Connection(connectionString);
    opts.AutoCreateSchemaObjects = AutoCreate.All;
});

// Wolverine (saga orchestration)
builder.Services.AddWolverineWithMarten(opts =>
{
    opts.PersistenceConnection(connectionString);
});

// Microsoft.Extensions.AI (LLM integration)
builder.Services.AddSingleton<IChatClient>(
    new OpenAIChatClient("gpt-4o", apiKey));

// Strategos
builder.Services.AddStrategos()
    .AddWorkflow<ProcessOrderWorkflow>()
    .AddThompsonSampling()
    .AddLoopDetection()
    .AddBudgetGuard();

---

Planned Integrations

pgvector

Status: Planned for Strategos.Rag

Will provide PgVectorAdapter implementing IVectorSearchAdapter for vector similarity search using the same PostgreSQL database.

Azure AI Search

Status: Planned for Strategos.Rag

Will provide AzureAISearchAdapter for enterprise-scale vector and hybrid search.

---

External Documentation

• Wolverine Documentation
• Marten Documentation
• PostgreSQL Documentation
• Microsoft.Extensions.AI Documentation