PulseStackAI

Modern .NET AI framework for building agents, tools, workflows, and enterprise AI applications.

Built on top of Microsoft.Extensions.AI, PulseStackAI provides a clean, extensible architecture for integrating AI into real-world .NET systems.

---

Why PulseStackAI?

PulseStackAI helps developers build production-ready AI systems without dealing with provider complexity, fragmented SDKs, or orchestration boilerplate.

Use it to create:

• AI copilots
• Multi-agent workflows
• Tool-calling assistants
• ERP / CRM AI integrations
• Enterprise knowledge assistants
• AI-powered automation systems

---

Framework Goal

PulseStackAI is designed to let developers focus on business orchestration rather than AI infrastructure.

The framework should abstract away:

• provider complexity
• orchestration coordination
• runtime execution concerns
• retries and resilience
• shared state management
• tool execution infrastructure
• execution diagnostics

while exposing clean, explicit orchestration primitives.

The goal is to allow developers to compose reliable AI workflows using deterministic runtime infrastructure.

---

Current Features (v0.2)

Providers

• OpenAI provider
• Unified IChatClient abstraction

Agents

• AgentBuilder fluent API
• Configurable instructions and temperature
• Streaming responses
• Structured tool calling

Tools

• Tool registry
• Dependency Injection integration
• Built-in tool support
• JSON-based tool execution loop

Pipelines

• Sequential multi-agent pipelines

---

Installation

Clone Repository

git clone https://github.com/abilgaiyan/PulseStackAI.git
cd PulseStackAI

---

Quick Start

Configure Services

services.AddPulseStack()
    .UseOpenAI(apiKey, "gpt-4o-mini");

---

Basic Chat

var client = sp.GetRequiredService<IChatClient>();

var response = await client.AskAsync(
    "Explain dependency injection.");

Console.WriteLine(response);

---

Build an Agent

var agent = new AgentBuilder("Assistant", client)
    .WithInstructions("You are concise and helpful.")
    .WithTemperature(0.3f)
    .Build();

var result = await agent.RunAsync(
    "Explain async/await.");

Console.WriteLine(result.Text);

---

Register Tools

services.AddPulseStack()
    .UseOpenAI(apiKey, "gpt-4o-mini")
    .AddTool<CalculatorTool>();

---

Tool-Enabled Agent

var registry = sp.GetRequiredService<IToolRegistry>();

var agent = new AgentBuilder("Assistant", client)
    .WithInstructions("Use tools when required.")
    .WithTools(registry)
    .Build();

var result = await agent.RunAsync(
    "What is 250 * 45?");

Console.WriteLine(result.Text);

---

Multi-Agent Pipeline

var researcher = new AgentBuilder("Researcher", client)
    .WithInstructions("Research the topic and provide key findings.")
    .Build();

var writer = new AgentBuilder("Writer", client)
    .WithInstructions("Convert findings into an executive summary.")
    .Build();

var pipeline = new AgentPipeline("ResearchPipeline")
    .AddAgent(researcher)
    .AddAgent(writer);

var result = await pipeline.RunAsync(
    "AI adoption in enterprise ERP systems.");

Console.WriteLine(result.FinalOutput);

---

Streaming Responses

await foreach (var chunk in agent.StreamAsync(
    "Explain dependency injection in simple terms."))
{
    Console.Write(chunk);
}

---

Solution Structure

src/
  PulseStack.Abstractions
  PulseStack.Core
  PulseStack.Agents
  PulseStack.Tools
  PulseStack.Providers.OpenAI

samples/
  BasicChat
  BasicAgent
  MultiAgentPipeline
  StreamingChat

tests/
  PulseStack.Tests

---

Architecture

Providers
    ↓
IChatClient
    ↓
Agents
    ↓
Tools
    ↓
Pipelines

---

Samples

Run Basic Chat

dotnet run --project samples/BasicChat

Run Basic Agent

dotnet run --project samples/BasicAgent

Run Multi-Agent Pipeline

dotnet run --project samples/MultiAgentPipeline

Run Streaming Chat

dotnet run --project samples/StreamingChat

---

Roadmap

v0.3

• Azure OpenAI provider
• Ollama provider
• Conversation memory
• Improved tool metadata
• Structured tool schemas

v0.4

• Vector storage
• RAG support
• Document indexing
• Semantic search

v1.0

• Enterprise workflow engine
• MCP integration
• Document intelligence
• AI governance / audit logging
• Plugin ecosystem

---

Current Status

PulseStackAI is currently in active early development.

The framework is stable enough for experimentation and prototyping while core architecture and APIs continue to evolve.

---

Design Philosophy

PulseStackAI focuses on:

• Clean Architecture
• Extensibility
• Provider abstraction
• Enterprise AI integration
• Developer productivity
• Incremental complexity

---

License

MIT License