Vitreus ChainForge
All docs

Getting started

From signup to a project that compiles.

Five steps. About fifteen minutes if you have an editor open and Rust on your machine; about thirty if you don't. By the end you'll have a working starter project on disk — a parachain skeleton you can extend, push, and eventually take to testnet.

Before you start — what ChainForge gives you

ChainForge generates a starter project that compiles. Not a tutorial, not a sandbox — a real parachain skeleton wired against Vitreus, with the conventions the chain expects already baked in. Your AI agent does most of the typing from there; ChainForge contributes the Vitreus-specific context so the agent produces code that compiles cleanly the first time.

What lands on your disk after five steps: a Cargo workspace, a custom pallet matching the template you picked, a runtime that wires it, a chain spec for genesis, a tuned CLAUDE.md your agent reads first, and a private GitHub repo that holds it all. Code on your machine, repo on your account, no ChainForge-side runtime path holding either of them. We're in the loop while you're building; out of it the moment you deploy.

What you bring — an AI coding agent

ChainForge produces the skeleton; an AI coding agent helps you flesh it out. Claude Code is the agent ChainForge tunes its conventions for — Anthropic's coding agent that runs in your terminal, reads your CLAUDE.md, and edits files in your repo as you ask for changes in plain English. Install at docs.anthropic.com/en/docs/claude-code.

Claude Code needs an Anthropic plan to run: either a Claude.ai Pro or Max subscription, or pay-as-you-go API credits. Current pricing is at anthropic.com/pricing.

Other AI coding agents work too. Cursor is the most common alternative — it reads CLAUDE.md as a project context file and applies the same conventions. GitHub Copilot and editor-bundled agents like JetBrains AI also work; anything that reads project-level instruction files can pick up the CLAUDE.md ChainForge writes.

If you don't have an agent set up yet, set one up before Step 1 — the rest of the guide assumes you have one.

Step 1 — describe your chain on the scaffold form

Open /scaffold and type what you want, in plain English. At least 20 characters; the more concrete the better — “a chain where users vote on which proposals to fund from a treasury” beats “a DAO platform.”

Below the description, pick a template or leave it on Auto:

  • Generic — minimum-viable starter, one custom pallet, no native token.
  • Token + Governance — native token plus on-chain voting (council + public referenda).
  • Prediction Market — yes/no outcome markets with stake-backed bets.
  • Marketplace — listings, offers, accept-or-cancel flow over generic fungible assets.

Auto routes you to the closest fit. The pallets the form mentions are the building blocks the generator wires underneath; you don't have to know what they are to submit, but the building guide covers it once you're ready to extend.

Step 2 — wait for the scaffold (about 15–30 seconds)

After you submit, the status pill walks four states: queuedgeneratingchecking ready. Generating is the LLM call: a model is reading your description and choosing pallet names, template constants, and the README intro. Checking is a sandboxed cargo check against the rendered scaffold — the platform compiles your project before handing it to you, so you don't download something broken.

Ready means both passed. If checking fails you see a clear failure card with the error and a Try again button — usually a one-word description tweak gets it through.

Step 3 — push to GitHub (or skip)

The success panel has a Push to GitHub button. Connect your GitHub account once (an OAuth handshake; the scope we ask for is repo, which lets us create a private repo on your behalf and push one initial commit). Click Push.

Private by default. Always. The repo's default branch is main, the initial commit is signed as “Initial scaffold from ChainForge,” and the topics include parachain and vitreus so it's findable on your dashboard later. If you'd rather keep the code off GitHub entirely, you can skip the push and download the scaffold directly — but Push is the smoothest path to step 4.

Step 4 — pull it down and open it in your editor

On the success panel, the View on GitHub button takes you to your fresh repo. The shape of the next bit depends on your tooling: you can clone it from the command line, or open it directly in an editor that handles GitHub URLs (VS Code, JetBrains, Cursor). To clone is to make a working copy of the repo on your machine — the standard command is git clone <repo-url>, run wherever you keep your projects.

Once it's open locally, run cargo check. Cargo is Rust's build tool; cargo check verifies your project compiles without producing a binary. The first run downloads the polkadot-sdk dependencies and takes a few minutes; subsequent runs are seconds. If it returns clean, you're ready for step 5.

If cargo check fails the first time, the most common cause is a Rust toolchain mismatch — the scaffold ships a rust-toolchain.toml that pins the channel; let rustup install it on first invocation. The reference guide's footguns section covers the other common failure modes.

Step 5 — read the CLAUDE.md

At the root of your project you'll find a CLAUDE.md file. It's a tuned brief addressed to the AI agent you'll work with — your Claude Code, Cursor, or whatever you're using. The agent reads it before touching anything else, picks up Vitreus-specific context, and produces code that matches the conventions ChainForge generated against.

You're welcome to read it yourself; it's plain markdown. But the prose is tuned for the agent — the “Your role here” section opens by addressing the agent directly, and the “Substrate / Vitreus footguns” section assumes the reader knows what a trait bound is. Skimming it gives you a useful mental model of what your agent will be optimising for; if you want the same content laid out for the human reader, the building guide covers the same patterns in plain English.

Where to go from here

You have a scaffold, it compiles, your agent has a brief. Three places to go next depending on what you want to do:

  • Ready to extend the pallet your agent generated? The building guide walks the working loop — how to prompt the agent, what conventions to expect in its output, how to verify before you push.
  • Curious what deploy looks like? The deploying guide covers what deploying a parachain involves — not the ChainForge walkthrough yet (that lands when the deploy walkthrough ships), but the conceptual shape plus the manual paths that work today.
  • Need to look something up? The reference guide carries the pallet catalogue, validator-rule lookup, footguns with their fixes, and a glossary if a term in your scaffold reads opaque.
For now: The deploy walkthrough that ships your chain to Vitreus testnet end-to-end is not yet live. For now, the deploying guide names every step and pairs it with the manual path you can drive today via polkadot.js apps plus the upstream substrate-node-template.