Use Cases Features Compare Atrium Blog GitHub
Talk to Us
All articles
forge ai-agents github 2 min read

Announcing Forge: From GitHub Issue to Pull Request with AI Agents

Introducing Forge — an AI-powered multi-agent platform where five specialized agents collaborate to turn GitHub issues into production-ready code.

by Ilia Sadykov

Today we’re announcing Forge — the AI-powered multi-agent development platform by Simple Container.

Most AI coding tools give you a single chat window. One model trying to do everything: analyze requirements, design architecture, write code, test it, and review it. That’s not how real engineering teams work.

Forge takes a different approach. It runs a pipeline of five specialized AI agents, each purpose-built for its role:

The Five-Agent Pipeline

  1. PM Agent — Receives the GitHub issue, analyzes requirements, and creates a clear development plan with acceptance criteria.

  2. Architect Agent — Takes the plan and designs the technical solution: component structure, API contracts, data models, and integration points.

  3. Developer Agent — Implements the solution following the architecture. Writes clean, tested code with proper error handling.

  4. QA Agent — Runs the test suite, validates edge cases, checks for security issues, and ensures the implementation matches the requirements.

  5. Code Reviewer — Reviews the final output against coding standards, best practices, and the original requirements.

How It Works

The workflow is intentionally simple:

  1. Label any GitHub issue with forge-pipeline
  2. Forge’s webhook picks it up and starts the pipeline
  3. Each agent hands off to the next with full context preservation
  4. Comments on the active issue are routed to the current agent
  5. The pipeline produces a reviewable pull request with a full audit trail

What’s Different

  • Deterministic orchestration: Through our LLMOS runtime, Forge provides deterministic control over AI workflows. Event-sourced state machines ensure reproducible outputs.
  • Multi-provider routing: Smart load balancing across Anthropic, OpenAI, Google, Azure, and Bedrock with automatic failover.
  • Cloud and local: Run locally with forge-cli for rapid iteration, or deploy the runtime on any machine — VMs, Kubernetes, GitHub Actions, or your own servers — with bidirectional sync.

We’re building Forge in the open. Read the Workflow Engine and Multi-Agent Pipeline pages, or browse the source on GitHub .