🧱 Laying the Foundation: Backend Bootstrapped, Monorepo Live

Stayed up late tonight to push something important over the line: the very first working version of the backend for Vulcan Forge. It’s not fully wired up yet, but the bones are in place.

I’ve now got a proper monorepo structure with three core components:

• vf-backend: The server powering the workflow. This is where users will create plans, prompts, nodes, blocks — all the ingredients of the build system.
• vf-frontend: Just the web UI for now, tied to the backend workflow.
• vf-anvil: This is the CLI utility (and eventually the desktop GUI) that actually reaches into the user’s codebase and makes edits. It’s the “forge” in action: where code gets reshaped.

What’s running today is vf-backend, bootstrapped mostly by porting over Simrata’s backend. It was the right call. It took a couple hours to migrate and clean up, but that’s a fraction of what it took to build it the first time. In Simrata, I probably spent a week or more getting auth, encryption, and job execution all humming. Now I can pull that same scaffolding in and get moving way faster.

It’s a great reminder: reusable modules like auth or job runners are no longer "nice to have." They’re leverage multipliers. They turn weeks into hours. Going forward, I need to keep thinking in terms of composable primitives that accelerate new builds, not just solve old problems. Building out primitives like this just got added to the backlog.

Right now, the backend passes lint, passes tests, and has models and structure in place. Routes still need setup. I haven’t tested login flows yet, but I’m close to being able to create and manage projects, blocks, and nodes.

Once that’s up, I’ll be able to use the workflow itself to start creating plans and prompts, run rubrics, and validate things into the database. The main code generation will still be manual via Cursor for now, but automating the plan and prompt audits and all that will be a real inflection point. The system will, for the first time, accelerate its own development.

What’s next:
Set up login and core CRUD routes. Once I can create blocks and nodes through the UI or API, I can begin wiring up the actual plan and prompt generation.