Documentation · Overview

From idea to your first pull request — using your Claude subscription

This is one continuous story: you have an idea, you have a Claude.ai subscription (Pro or Max), and you want CogniForge to design and build the first version of your product with it — at no extra AI cost. No prior technical experience assumed. By the end you'll have reviewed and merged a real pull request into your own repository.

The one decision you make up front: everything CogniForge does for you — writing your project's plan, and later writing code — needs one machine to actually run on. Call it your engine. You have two choices, and every step on THIS page is identical either way — only how you set up your engine and coding workspace (steps 2 and 5 below) differs, and each has its own dedicated page so you never have to sort through instructions meant for the other path.

1Create your account, describe your idea, and create your project

1a. Create your account

  1. Go to your CogniForge platform's web address and click Sign up.
  2. Fill in: Company name (anything — this becomes your own private workspace, called a "tenant"), your name, your email, and a password (at least 8 characters).
  3. Submit — you're signed in immediately, and you are automatically the administrator of your new workspace. Nobody else can see it unless you invite them.

1b. Describe your idea and create your project

  1. Click New project, choose Greenfield (meaning: building something new, not touching existing code).
  2. Give it a name, and write a paragraph or two about what you want to build in plain English — who it's for, what they'll be able to do with it, what problem it solves. Don't worry about being complete or technical; you'll be asked clarifying questions next.
  3. Submit — your project is created.
CogniForge immediately tries to write your first planning document — but nothing can connect to your Claude subscription yet (that's step 2, next), so this first attempt is expected to just sit on a spinner. That's normal, not an error: once step 2 is done, you'll click one button to pick it back up (see step 3).

2Set up your engine, powered by your Claude subscription

This is the step that makes your subscription actually power CogniForge — both the planning documents in step 3 and, later, the code itself. Pick whichever of these matches your situation; each page is a complete, self-contained walkthrough for that path only, and will send you back here (to step 3) when it's done.

3Review and approve your project's plan

Go back to your project. If it's still sitting on the spinner from step 1, look underneath it for a Retry generation button and click it — this is the expected, normal way to pick generation back up now that step 2 is done, not a sign anything went wrong. Within a minute, an Understanding Document should appear — CogniForge's expansion of your idea into personas, capabilities, and a proposed shape for the product, ending with a section called Assumptions & Open Questions.

  1. Read the whole thing, especially the open questions.
  2. Type answers into the Refine box in plain English (e.g. "yes, users pay monthly, no free tier") — the document updates itself, only in the parts your feedback touches.
  3. When it matches what's in your head, click Approve. This automatically starts generating the next document, the Constitution (the non-negotiable engineering rules) — review and approve it the same way.
  4. Then the Solution Architecture Document generates — the real technical design: what languages/frameworks it'll use, what "done" means for each piece (the Testing Bar), what it'll look like (design system), and the milestone plan. Read it, refine anything you have an opinion on (even "I'd prefer X over Y" in plain English), and approve.
Approving the Solution Architecture Document is the moment your project's board appears — a real backlog of tickets, sized as complete pieces of work, Milestone 1 first.

4Create and connect an empty repository

Your code needs somewhere to live. If you don't already have a GitHub account, create one free at github.com.

  1. On GitHub, click +New repository. Give it a name, leave everything else at its default (don't add a README), click Create repository. Leave it empty — CogniForge makes the first commit itself.
  2. Still on GitHub, go to your account's Settings → Developer settings → Personal access tokens → Fine-grained tokens → Generate new token. Give it access to just the repository you created, with permissions Contents: Read and write and Pull requests: Read and write. Generate it and copy the token — GitHub only shows it once.
  3. Back in CogniForge, open your project's Repositories tab → Connect repository. Paste the repository's URL and the token.

CogniForge clones it on your engine machine and starts building its knowledge index — nothing for you to do here but wait for the status to say Ready.

Now go set up your coding workspace — the second half of whichever path you picked in step 2 (local laptop, Part 2 or IDE local + VM, Part 2) — then come back here for step 5.

5Point tickets at your workspace

  1. In your project, go to Settings → Auto-Deliver, and change the delivery option from "Platform (build VM)" to "Connected IDE (Claude Code)". This tells CogniForge to send tickets to the VS Code window you just connected, instead of building them invisibly on the server.
Keep VS Code open (or your VM's connection alive) whenever you want tickets to be worked — CogniForge delivers to whichever connected workspace is currently online.

6Start building, and watch it happen

  1. Open your project's Board and click Start auto-deliver, then confirm.
  2. Watch a ticket move to In Progress — in VS Code, you'll see the CogniForge sidebar say "Working on…", and a headless Claude Code session runs in that folder, using your subscription, implementing that ticket's scope, running its tests, and pushing its commits.
  3. One by one, tickets complete and the next one starts automatically. When every ticket in Milestone 1 is done and the whole system passes its end-to-end tests, CogniForge opens a pull request with everything, and auto-deliver pauses itself waiting for you.

7Review and merge your first pull request

  1. Open the pull request link CogniForge shows you (or find it on your GitHub repository's Pull requests tab).
  2. Skim what changed — CogniForge's summary explains it in plain terms. Everything in it already passed real build and test runs; nothing landed here that didn't work.
  3. Click Merge pull request on GitHub.
  4. Back in CogniForge, go to the Milestones page and click Approve (it only lights up once every ticket is Done). This isn't just a sign-off — it confirms the PR actually merged, waits for the merged code to finish re-indexing, then runs an automated technical-debt review (using this same Claude subscription) against what you just shipped. A clean result approves the milestone and unlocks Milestone 2. If it finds something worth fixing, it files new tickets on Milestone 1 instead and tells you so — let auto-deliver work through those, then click Approve again.
  5. Once approved, click Resume auto-deliver to keep going with Milestone 2, whenever you're ready.
You just shipped your first milestone — designed, built, and tested using nothing but your own idea and your existing Claude subscription. Everything after this point (Milestone 2, 3, …) repeats steps 6–7 only (including the Approve review each time); steps 1–5 were one-time setup.
From idea to your first pull request, using your Claude subscription — CogniForge Docs