Build an AI code review team with .NET Microsoft Agent Framework background agents and Ollama

Authorised Territory demonstrates how to build a multi-agent AI code review workflow in .NET using Microsoft Agent Framework, coordinating specialist security, performance, and architecture reviewers via background agents. The setup runs fully locally with Ollama and adds OpenTelemetry tracing so you can see what each agent is doing during the review.

Overview

The video builds an AI-powered “code review team” where multiple specialised agents independently review an ASP.NET Core MVC application from different perspectives, and a lead agent consolidates the results into a single structured report.

Key ideas covered:

• Use Microsoft Agent Framework in a .NET console app to orchestrate multiple agents.
• Run the LLM locally using Ollama with the granite4.1:8b model.
• Connect to Ollama through an OpenAI-compatible local endpoint.
• Coordinate concurrent specialist reviews using BackgroundAgentsProvider.
• Add OpenTelemetry tracing (including console exporters) to observe agent activity.

Multi-agent code review design

Roles in the review team

The workflow is structured as:

• User → Lead Developer Agent → Specialist Agents

Specialist agents:

• Security Agent
  • Looks for issues such as:
    • SQL injection
    • Authentication and authorization problems
    • XSS
    • CSRF
    • Secrets and sensitive data exposure
    • Insecure model binding
• Performance Agent
  • Looks for issues such as:
    • N+1 queries
    • Inefficient LINQ
    • Unnecessary database round trips
    • Excessive entity loading
    • Missing AsNoTracking()
    • Synchronous operations
• Architecture Agent
  • Evaluates:
    • Separation of concerns
    • Controller responsibilities
    • Dependency boundaries
    • Business logic placement
    • Data access patterns
    • Testability and maintainability

Lead Developer Agent responsibilities

The Lead Developer Agent coordinates the review using Microsoft Agent Framework’s BackgroundAgentsProvider:

• Delegates tasks to specialist agents
• Runs specialist reviews independently and concurrently
• Combines results into a single report by:
  • Removing duplicates
  • Resolving conflicts
  • Prioritising issues
  • Producing a structured final code review output

Guardrails for specialist agents

The specialist agents are instructed to:

• Focus strictly on their assigned responsibility (security/performance/architecture)
• Treat the supplied source code as the authoritative source
• Avoid inventing files, classes, methods, or vulnerabilities

Implementation topics highlighted

The video’s build steps and features include:

• Building a .NET console application with Microsoft Agent Framework
• Connecting Microsoft Agent Framework to Ollama
• Configuring an OpenAI-compatible local endpoint
• Creating specialised AI agents
• Using BackgroundAgentsProvider to coordinate background agents
• Building a lead developer agent
• Delegating tasks to specialist agents
• Running independent reviews concurrently
• Combining multiple AI reviews into a final report
• Using LoopAgent and streaming responses
• Adding OpenTelemetry tracing and console exporters

Why run locally with Ollama

The approach keeps the LLM running locally (via Ollama), enabling experimentation with multi-agent code review workflows without relying on a cloud-hosted model.