Launched this week
ArchFlow
Open-source AI architect: from requirements to code
11 followers
Open-source AI architect: from requirements to code
11 followers
ArchFlow is an open-source, self-hosted AI system architect. Import a PRD or meeting transcript and get traceable requirements, AI-proposed architectures you compare and record as ADRs, and low-level designs for every component: APIs, tables, events and diagrams. Check health, cost and compliance (SOC 2, HIPAA, GDPR, PCI DSS), then generate Terraform, Kubernetes, tests and specs, and hand the design to Claude Code and other agents over MCP. Your team edits it live, with branches and reviews.
Free
Launch Team
CloudflareYour next app belongs on Cloudflare Workers. Deploy for free
Promoted
Hi Product Hunt
I'm Narender, the maker of ArchFlow.
Why I built it: on every team I've worked with, system design lives in scattered places: a requirements doc, a few diagrams, ADRs nobody updates, and code that has quietly drifted away from all of them. AI made writing code faster, but the architecture behind it is still guesswork. I wanted one place where requirements, architecture, decisions and code stay connected, and where people and AI agents work on the same model instead of copying things between tools.
What ArchFlow does:
📄 Requirements: import a PRD or a meeting transcript; ArchFlow extracts requirements and traces each one to the components that implement it.
🧭 High-level design: the AI architect asks a few questions, then proposes complete architectures with cost, pros and cons. Open them as draft branches.
⚖️ Decisions: compare drafts against weighted priorities and record an ADR linked to the requirements and components.
🔬 Low-level design: open any component to design its internals (API endpoints, tables, events), with C4, sequence, ER and deployment diagrams.
✅ Production checks: a health score, AI review agents, a cost estimate, and SOC 2 / HIPAA / GDPR / PCI DSS / ISO 27001 checks.
⚙️ To code: generate Terraform, Kubernetes, tests, SRS and runbooks, then hand the design to Claude Code and other agents through OpenSpec and MCP.
🔁 Code back to design: rebuild the architecture from an existing repo, and see drift and change impact.
👥 Team: live collaboration, comments, branches, merges and approvals.
Open source and self-hosted. Bring any model through LiteLLM (Gemini, Claude, OpenAI, local models via Ollama, and more), and the app runs fine without AI as well.
🎬 A full walkthrough designing a global payments platform end to end: https://youtu.be/hqfFbIlfeHc
⭐ GitHub: https://github.com/narender-rk10/ArchFlow
I'd love your honest feedback, especially:
1. Which part of system design is most painful for your team today?
2. What would you want ArchFlow to generate or connect to next?