FlowForge

What it does

FlowForge transforms natural language descriptions into production-ready GitLab Duo custom flows and agents. Describe your desired DevOps automation in plain English — FlowForge generates valid flow YAML, agent configurations, and trigger setups that deploy directly to the GitLab Duo Agent Platform.

## How we built it

• Frontend: Next.js 16 with React 19, Tailwind CSS 4, Monaco Editor for YAML editing, Framer Motion for animations
• AI Engine: Anthropic Claude SDK for NL→YAML generation with real-time streaming
• GitLab Integration: 2 custom public agents (FlowForge + Security Sentinel) created on GitLab Duo Agent Platform, 2 custom flow YAMLs, GitLab API v4 for deployment
• Deployment: Vercel with live demo

## GitLab Duo Agent Platform Usage FlowForge includes two custom public agents that TAKE ACTION autonomously:

• FlowForge Agent — when mentioned in issues/MRs, generates custom flow YAML and creates issues with deployment instructions
• Security Sentinel — scans code for vulnerabilities, creates issues with severity labels, generates merge requests with fixes

Both agents use Claude via GitLab AI Gateway ($AI_FLOW_AI_GATEWAY_TOKEN) — no personal API key required.

## Challenges we ran into The GitLab Duo custom flow YAML schema is partially documented, requiring reverse-engineering from built-in flows and the flow registry v1 specification. We solved this by building a YAML validation layer that checks against known schema constraints.

## What we learned The GitLab Duo Agent Platform is incredibly powerful for automating DevSecOps workflows. The combination of custom agents, flows, triggers, and the AI Gateway creates a platform where AI agents can take real autonomous action — not just chat.

## What's next for FlowForge

• Visual drag-and-drop flow editor
• Flow versioning and rollback
• Multi-agent orchestration with conditional routing
• Integration with GitLab CI/CD for automated flow testing before deployment

Built With

• css
• next.js
• react
• tailwind
• typescript