AutoForge

• 

  AutoForge — Multi-Agent DevOps AI System 3 Agents • 3 Flows • Native GitLab Execution

• 

  SRE Agent Auto-Fix Diagnoses and fixes CI/CD failures automatically

• 

  GreenOps Intelligence Tracks carbon footprint and optimizes pipelines

• 

  Security Agent Detection Identified critical vulnerabilities automatically

Inspiration

DevOps teams spend over 40% of their time on reactive work — debugging CI/CD failures, fixing vulnerabilities, and manually reviewing pipelines. These are repetitive, pattern-heavy tasks, making them ideal for automation.

Most AI tools today act as assistants — chatbots that suggest fixes but don’t take action. We wanted to go further:

What if AI could act like a team of engineers — investigating problems, making decisions, and fixing issues autonomously inside GitLab?

With the GitLab Duo Agent Platform, this became possible. AutoForge embeds AI agents directly into the development workflow — not as external tools, but as active participants.

We also introduced a GreenOps perspective, proving that CI/CD pipelines can be measured and optimized for sustainability, not just performance.

---

What it does

AutoForge transforms GitLab into a self-healing DevOps system using three specialized AI agents:

🛠 SRE Agent

• Detects and analyzes CI/CD pipeline failures
  
• Uses structured reasoning (Tree-of-Thought) to identify root causes
  
• Automatically creates fix merge requests
  

🔐 Security Agent

• Scans code for vulnerabilities (OWASP, CVEs, injection flaws)
  
• Assesses severity and generates patches
  
• Creates issues and remediation fixes
  

🌱 GreenOps Agent

• Measures pipeline energy consumption using real calculations
  
• Estimates carbon footprint (kWh → CO₂)
  
• Suggests optimizations for efficiency and sustainability
  

These agents are orchestrated through flows, which allow them to:

• React to triggers (@mentions, assignments)
• Coordinate multi-step reasoning
• Execute actions autonomously inside GitLab
  

---

Why this is different

AutoForge is not just another AI tool — it acts, not assists:

• No chat prompts → Triggered by real GitLab events
  
• No suggestions → Creates real merge requests & issues
  
• No external tools → Runs entirely inside GitLab
  
• Not single-agent → Coordinated multi-agent system
  

This shifts DevOps from:

❌ Manual debugging → ✅ Autonomous resolution

---

How we built it

AutoForge is built entirely on the GitLab Duo Agent Platform using:

• Custom AI agents defined in YAML
  
• Multi-step orchestration flows
  
• Built-in Anthropic Claude models
  
• Native GitLab tools (MRs, issues, pipelines)
  

Each agent follows a 5-phase cognitive pipeline: Perceive → Reason → Plan → Act → Reflect

This allows structured, explainable decision-making instead of simple prompt-response behavior.

---

Challenges we ran into

• GitLab hackathon CI restrictions required adapting to trigger-based flows
  
• Limited documentation for Duo GraphQL APIs required experimentation
  
• Designing consistent structured reasoning outputs (Tree-of-Thought)
  
• Balancing accuracy vs simplicity in GreenOps energy calculations
  

---

Accomplishments that we're proud of

• Built a complete multi-agent system with 3 agents + 3 flows
  
• Successfully triggered real workflows inside GitLab
  
• Generated actual merge requests, issues, and pipeline fixes
  
• Implemented sustainability tracking using real energy models
  
• Achieved full native integration with GitLab Duo (no external infra)
  

---

What we learned

We learned that AI becomes truly powerful when it moves beyond assistance into execution.

By embedding agents directly inside developer workflows, we enabled AI to:

• Understand context
  
• Make decisions
  
• Take real actions
  

This represents a shift toward autonomous software engineering systems.

---

What's next

• Learning loops using past fixes
  
• Cross-project intelligence
  
• Human-in-the-loop approvals
  
• Additional agents (QA, Code Review, Docs)
  
• Carbon analytics dashboard
  

AutoForge is just the beginning of self-operating DevOps systems.

Built With

• anthropic-claude-(via-gitlab-duo)
• gitlab-ai-catalog-(flow-registry-v1)
• gitlab-ci/cd
• gitlab-duo-agent-platform
• multi-agent
• python
• tree-of-thought-reasoning
• yaml