Copied tegeljb@gmail.com to clipboard
← All docs

Using Forge

Getting Started

This walks through what actually happens after you’ve installed Forge (see Getting Started if you haven’t yet) and open it for the first time.

Describe your first project

Type what you want in plain words — “an app for my dog-walking business,” “a tool that tracks my expenses.” Forge picks a real project template for that kind of app and generates the first working files. You don’t need to know what a “template” means technically; Forge handles the technical decisions and asks you the questions that matter (does it need logins? what happens if something fails?) instead of leaving you at a blank prompt.

The stage hint tells you what’s next

Once a project exists, a hint bar above the chat tells you where your project actually is, based on scanning the real code — not a guess:

  • Sketch — nothing built yet. Describe what you want below.
  • Skeleton — files exist but barely any functions/classes yet. This is when Forge suggests trying 🧭 Analyze Gaps.
  • Flesh — there’s a real amount of code. Tell Forge what you’re adding or fixing next.
  • Skin — the project has tests and documented functions. Good time to review or prepare a release.

This updates automatically as your project grows, including in the background every 15 minutes while Forge is open, so it never goes stale.

Analyze Gaps

Click 🧭 Analyze Gaps any time — it checks what you’ve described (or the folder you have open) for missing pieces: tech stack decisions, who the app is for, whether there’s any testing plan. It’s a good habit right after the first scaffold, but you can run it whenever something feels incomplete.

Show my plan

As Forge learns more about your project, it keeps a running plan behind the scenes — not just a chat history, but an actual structured breakdown (overview, architecture, features, data, implementation steps, risks, user experience). Click 📋 Show my plan any time to see it. This is also what makes picking a project back up later easier: the plan persists, so if you close Forge and come back tomorrow, it’s not starting from zero.

Run generated code (sandboxed)

Forge never runs code on your machine without you asking. When you want to actually test something it wrote, click 🧪 Run generated code (sandboxed) — this runs it in a real isolated environment (not just a warning label) that can’t touch files outside your project folder, can’t reach the network, and can’t spawn other programs. If it hits an error, you’ll see exactly what happened; nothing runs silently in the background.

Add a feature using a template

Already have a project going and want to add something new to it — like authentication, or a database — rather than starting over? Click + Add a feature using a template to reuse the same guided template flow on your existing project instead of a blank one.

Check Licenses

If your project pulls in third-party code or libraries, this checks their licenses so you’re not caught off guard by a restriction you didn’t know about — worth a look before you consider a project “done,” especially if you plan to share or sell it.

Cloud providers (optional)

Everything above works with the free local Ollama engine, no account or key needed. If a task needs more capability than the local model gives you, Forge will tell you and offer to escalate — at that point you can add your own key for one of: Gemini (free tier, recommended default), Groq, DeepSeek, Anthropic Claude, or Ollama Cloud, under Settings → API Keys. You only need one, and you’re never required to add any of them to use Forge.

Going further with an AI coding console

Forge scaffolds a real starting point — picks a template, writes the first working files, checks them. If you want a full project built end-to-end with an AI doing the architecture, pair Forge with a coding console pointed at the folder Forge created. Forge isn’t tied to one of these — it exposes its project context (plan, rules, file state) over MCP, an open standard several consoles already speak, not a Forge-specific format.

Claude Code (Anthropic’s official coding CLI) gets the richest integration: Forge writes the connection into your project automatically, and Claude Code also picks up Forge’s turn-by-turn hooks, so Forge’s own view of your project (which files exist, what changed) stays current without you doing anything. Requires a Claude Pro, Max, Team, Enterprise, or Console account (not included in the free Claude.ai plan) — a separate, optional tool with its own account, not part of Forge itself.

GitHub Copilot (in VS Code’s agent mode) reads the same connection Forge already wrote for your project — no extra setup. It gets Forge’s project context (plan, rules, known files) whenever it asks for it, but not the automatic turn-by-turn tracking Claude Code gets, since that part is specific to Claude Code.

Other MCP-aware consoles (for example OpenAI’s Codex CLI, or Cursor) can connect too, using the same server — we just haven’t tested each one ourselves yet, so treat this as “should work,” not “confirmed.” Forge already wrote the connection details into your project’s .mcp.json file; most consoles let you add an MCP server by pointing at that same command, usually through a settings screen or a command like <console> mcp add. If you try one and hit a snag, tell us — we’ll test it properly and correct this page.

Whichever console you pick, expect the same shape: it can read what Forge knows about your project when it asks, but only Claude Code currently updates that picture automatically as you work.