DEVFORGE

Devforge is a desktop app for Windows and macOS that sits between you and Claude Code (or Aider, Codex CLI, Gemini CLI). You describe what you want built. Devforge assembles a structured prompt with your full project context, your design document, your task list, your session history, and sends it to the AI. The AI writes code in your project folder. You review the results.

If you don't know what game you want to make, this won't help you. Devforge doesn't generate ideas. It doesn't make creative decisions. You bring the design, the taste, the vision. The AI handles the typing.

How it works

You add a project, point it at a folder, and pick a mode. There are 17 modes. FORGE turns a game concept into a structured GDD, task list, and project files. IMPLEMENT translates your designs into working code. DEBUG does root cause analysis. PITCH stress-tests your ideas before you commit to building them. LENS reviews your content for cultural representation, content rating, and regional sensitivity concerns. Each mode gives the AI a different role and set of instructions tailored to that job.

Devforge reads your project files before every prompt. Your GDD, your task list, your session notes from last time. The AI starts each conversation already knowing your project, your rules, your current phase. You don't re-explain anything.

What's in the box

• 17 specialized modes (Forge, GDD, Pitch, Discuss, Implement, Debug, Research, QA, Freeform, Marketing, Instruct, Test, Security, UI/UX, Mod, Package, Lens)
• 20 platform stacks (Godot, Unity, Unreal, Phaser, PICO-8, GB Studio, Love2D, plus Swift/iOS, Kotlin/Android, Ruby/RPG Maker, asm6502/NES, Ikemen GO, and more)
• 40+ game dev skills injected into prompts based on your stack
• Analog mode toggle for board game and tabletop design
• Git safety: session branches, safety snapshots, one-click undo
• Parallel terminal tabs with git worktree isolation
• Local AI support via Ollama for free utility tasks (summaries, session notes, prompt expansion)
• Built-in playtest logger that routes observations to Debug, Discuss, your task list, or generates a structured markdown playtest report
• Session continuity across work sessions
• Cost transparency: pre-send dollar estimate, live running-cost chip during streaming, burn-rate alert on runaway responses, daily/weekly budget caps, and a STATS panel with per-project, per-mode, and Top Spend breakdowns
• ASK Bot pop-out: Ollama-powered help bot that lives in its own window, keeps live context while Claude works, and stays visible across mode switches
• Native notifications via Windows Toast / NSUserNotification, pinged when long runs finish even when minimized
• Anti-slop filter for all human-facing copy (Marketing mode, Share, AI Notes). A 10-rule system that strips filler, passive voice, false agency, banned AI words, and formulaic structures from any prose the AI generates. Based on stop-slop by Hardik Pandya. Your devlogs, store descriptions, and session summaries read like a person wrote them.
• CRT dark theme and SNES light theme

Who this is for

You've been designing games in notebooks, spreadsheets, or docs. You have a GDD on your hard drive that stalled because you hit a technical wall. You know what you want but you don't want to write the code yourself, or you'd rather spend your time on design than implementation. You've used Claude Code or similar tools but got tired of re-explaining your project every session.

Who this is NOT for

You want AI to come up with your game idea. You want to type "make me a platformer" and get a finished game. You don't have a clear vision for what you're building. Devforge will frustrate you.

Requirements

• Windows 10/11 or macOS 14+ (Apple Silicon)
• Claude Code CLI installed (or Aider, Codex CLI, Gemini CLI)
• Optional: Ollama for free local AI features

Same applies to the devlog post body — itch's devlog editor is the same rich-text-or-HTML setup. HTML version of the v0.3.1 devlog:

v0.3.1 ships three big things, plus a stack of polish.

Mac support. Apple Silicon Macs can run Devforge now. The .dmg is on the downloads page alongside the existing Windows installer.

Cost transparency. Built-in cost tracking shows what each Claude session is actually costing you. Pre-send dollar estimate, live running-cost chip while Claude works, burn-rate alert on runaway responses, daily and weekly budget caps. Computed locally. No data leaves your machine.

LENS mode (the 17th). Editorial and sensitivity reviewer. Stress-tests how your content will be perceived: cultural parallels, content rating implications (ESRB / PEGI / CERO), regional sensitivity, language audit. Not a censor: flags concerns with severity tiers and 2-3 options per finding. You decide what to do with each one.

Plus: TEST mode REPORT button (auto-grouped markdown playtest reports), HABITS dashboard (your last-7-days mode usage and spend patterns, computed locally), ASK Bot pop-out (the Ollama-powered helper now lives in its own window so it stays visible across mode switches), Code Browser language coverage expanded from 14 to 20 stacks, and native OS notifications.

Full changelog in the download zip.