> DEVFORGE

A game dev terminal that knows your project cold.

Your GDD stays loaded. Claude reads it before every session, along with your stack, your tasks, and last session's notes.

v0.4.0 live · $9 early access · No subscription · Windows + macOS

Buy on Itch.io // $9

What DevForge Is

DevForge is a structured UI that wraps AI coding tools. It's not a game engine and it's not going to design your game. You make the design decisions. DevForge handles context, modes, and version control so Claude stays focused on your project.

Features

Your GDD Runs the Show

Claude reads your design doc before every prompt. Your vision stays intact whether you're writing code, testing balance, or drafting store copy.

17 Modes, 17 Jobs

DISCUSS talks design. IMPLEMENT writes code. PITCH stress-tests your scope. LENS stress-tests how your content will be perceived. MOD builds mods. PACKAGE handles distribution. Each mode locks Claude to one job.

Git Safety Net

Session branches, safety snapshots, one-click undo. Built for people who don't touch git and don't want to learn.

Nothing Gets Lost

Your session notes, task progress, project rules, and GDD sections ship with each prompt. You pick up exactly where you stopped.

Claude Learns Your Rules

After each session, Ollama reads the transcript and suggests project rules: corrections, conventions, gotchas you hit during the session. Accept, edit, or reject each one. Claude reads them next session.

Watch Claude Work

A live activity feed shows every file read, edit, and bash command. Toggle the thinking display to see Claude's reasoning. Press Escape to stop mid-execution.

Smart Model Routing

DevForge picks the right Claude model per task. Opus for architecture and debugging. Sonnet for implementation. Haiku for quick lookups. Saves subscription time on simple tasks, uses full power when it matters. Toggle it off and pick manually.

Two themes.

SNES Light
CRT Dark

17 modes. One job each.

Analog Mode for tabletop design.

Toggle Analog in the footer and all 15 modes shift to tabletop. IMPLEMENT writes rules text and component specs instead of code. DEBUG finds rules contradictions. QA audits component consistency and first-play clarity. LENS reviews faction naming, historical framing, and gendered rule language. MARKETING writes Kickstarter pages and BoardGameGeek listings.

Design a board game with Analog on, then flip it off and use Implement to build the digital version. Your GDD carries over.

START. Work. END.

START

Creates a safety snapshot and optionally a session branch. If Ollama is running, it summarizes your project state so Claude starts oriented.

Work

Pick a mode, type what you want, send. Your project context ships with each prompt. The diff viewer, task panel, and git status refresh after each response.

END

Review all changes in the diff viewer. Write session notes (free via Ollama, or use Claude). Undo the entire session back to the snapshot if anything went wrong.

Up to 4 sessions at once.

Click + on the terminal tab bar to open a parallel session. Each tab gets its own git worktree, its own branch, its own Claude session with resume. You work on multiple tasks and they stay isolated. When you're done, MERGE folds changes back into your main branch. CLOSE discards.

Capped at 4 tabs since each one burns tokens independently. Tab switching is instant. Each terminal keeps its full scrollback.

There's also a separate BUILD/TEST terminal (toggle with the BUILD button) that runs your project's build and test commands apart from Claude's output. Stack-aware: cargo build for Rust, npm run build for HTML/Phaser, and so on.

Built-in tools.

Live Research Tools

Research mode searches the live web, downloads and summarizes books, and checks what's trending in game dev. SETUP RESEARCH walks you through installing the tools. DevForge saves your findings to the knowledge bank and includes them in future prompts. Optional, free.

Anti-Slop Filter

10-rule post-processing system applied to share summaries, Marketing output, and AI-generated session notes. Strips em dashes, filler phrases, passive voice, and false agency. Your writing sounds like you wrote it.

Ollama (Free Local AI)

Optional. Runs locally so you don't burn provider tokens: AI Notes, smart context briefings, prompt expansion, smart routing, diff descriptions, GDD suggestions, project rule extraction, and an ASK bot with live visibility into Claude's current task. Ask "what is Claude working on?" mid-session and get a real answer. Everything works without Ollama. You just lose the free stuff.

50 skills. Injected when relevant.

37 digital: Roguelike architecture, platformer mechanics, state machines, collision detection, camera systems, pathfinding, dialogue trees, save/load, inventory, animation, audio, particles, shaders, OWASP security auditing, WCAG accessibility, visual diagrams & charts, Godot C# patterns, plus stack-specific patterns for all 20 platforms.

13 tabletop: Rules writing, component design, balance & probability, hex maps, area movement, point-to-point maps, solitaire bot systems, wargame scale, VASSAL modules, Tabletop Simulator mods, print & play layout, Nandeck card layout.

20 platforms.

Godot 4 (GDScript), Godot 4 C# (GodotSharp), Unity, Unreal, Rust, Python, Love2D, GameMaker, RPG Maker, Phaser 3, HTML5, PICO-8, GB Studio, NES, Game Boy, Genesis, GBA, Ikemen GO/MUGEN, iOS, Android. There's also an "Undecided" option — FORGE asks about your game and recommends the right stack. DevForge injects stack-specific build commands, docs, and coding conventions into every prompt.

4 providers. Auto-detected at startup.

DevForge checks which CLIs are installed when you launch it. Green dot means ready. Red dot means missing, with install instructions right there. No digging through docs.

Claude Code ($20-200/mo) is the tested, recommended provider. Session resume, streaming output, direct file editing, context tracking, cost reporting. Smart model routing picks Opus, Sonnet, or Haiku per task.

Gemini CLI (free) runs with full activity feed, session resume, and auto-approval. Same experience as Claude Code. Use Google AI Studio's free tier to get started with no credit card.

Aider (bring your own key) and Codex CLI (OpenAI credits) are configured and the architecture supports them. For providers without session memory, DevForge injects heavier context: session notes, conversation log, GDD sections, code map.

A built-in PROVIDERS help tab covers install commands, authentication, costs, and troubleshooting for every provider. Switch providers and get a connection test confirming it works.

Looking for testers willing to try DevForge with Aider or Codex and report back. Free access for multi-provider testers.

Windows and macOS today. Linux maybe later.

DevForge ships on Windows 10/11 and macOS 14+ (Apple Silicon). The Mac build went through a full smoke pass before launch — provider auto-detect, mode switching, theme toggle, end-to-end Claude session, all green. Linux hasn't been tested but the codebase has nothing platform-specific blocking it; Tauri 2 builds for all three.

On Linux and willing to test? Free access for platform testers who report bugs.

Everything else.

  • Voice Profile Paste writing samples into Marketing mode. You get output in your voice.
  • Code Map Auto-generated file structure summary. Claude doesn't waste tokens exploring your repo.
  • Phase Tracker Visual progress bar parsed from your task list checkboxes.
  • Scope Meter Auto-calculated from lines of code, file count, git commits, GDD size, tasks, and stack. Ships with every prompt as a guardrail. Switches to manual target mode in Forge.
  • Knowledge Banks Per-project storage for Research, QA, Marketing, and Instruct, plus a global cross-project bank.
  • PDF Export One-pagers, press kits, and save-as-PDF from Marketing mode.
  • Session Transcripts Every Claude response saved as markdown. Browse, rename, filter, or delete past transcripts. Load any transcript as context for your next prompt.
  • Diagram Generation Quick actions in 8 modes to generate architecture diagrams, flowcharts, infographics, and comparison charts as PNG. Mention "diagram" or "chart" in any task and the render skill auto-injects.
  • Cost Tracking Footer shows "WORK DONE: $X.XX" with color-coded daily spend. Per-project totals so you see which games get the most AI investment. DevForge tracks usage against your existing AI subscription. No separate billing.
  • /CONTEXT Live session state: model, CLI version, context window usage bar (color-coded), token breakdown including cache hits, session cost.
  • /USAGE Today's and last 7 days' cost, tokens, sessions, daily average. Powered by ccusage.
  • 243 Automated Tests Provider spawning, argument safety, dual-path consistency, security encoding. 9 patches from external tester feedback.
  • CLI Resilience DevForge detects your provider's CLI version at startup. If an update introduces breaking changes, you get a warning in the log panel instead of silent failures.
  • Two Themes CRT Dark (retro terminal) and SNES Light (Super Nintendo palette).
  • Oryx Sprites + SFX 43 pixel art characters with idle, running, success, and error animations, plus 6 boss sprites for milestone events. 8-bit sound effects. All toggleable.

What's next.

Just Shipped — v0.4.0

  • 103 New Stack Skills Platform coverage now 20+ engines and targets: iOS/Android, Godot, Unity, Unreal, Ren'Py, LOVE, GameMaker, RPG Maker, GB Studio, NES/GB/GBA/Genesis, PICO-8, plus 10 classic source-port MOD targets (Doom, Quake 1-3, Diablo, Duke 3D and friends). Every skill web-verified against current docs before shipping.
  • Run Coordinator Sidebar shows live status of all active AUTO tabs, with crash recovery.
  • hands_off.yaml Awareness AUTO tabs warn before touching files a project marked hand-tuned.
  • Per-Provider Concurrency Caps Claude 4 / Gemini 3 / Ollama 2, global 9. Replaces the flat limit.
  • Per-Tab Provider Runtime AUTO-tab resume fixed end-to-end. MERGE action for parallel worktree tabs.
  • Security Hardening Closed an arbitrary-executable hole in session spawning with a hard provider allowlist. SSRF allowlist hardening with opt-in remote Ollama hosts.
  • Obsidian Vault Export Export your project's GDD, session notes, transcripts, research, and task lists from the project sidebar into a linked Obsidian wiki.
  • Smaller Downloads Thin-LTO + stripping. Smoke suite grew from 496 to 1085 checks, all green.
  • Docs Rewrite MANUAL and QUICKSTART rewritten to cover the autonomy arc.

Coming Next

  • Smarter Autonomous Tabs Ollama reads your task list and feeds tasks to Claude one at a time. Autonomous tabs become multi-step workflows instead of single prompts. Progress reports after each step.
  • Cross-Provider Routing Route different types of work to different providers. Claude for code, Gemini for research, Ollama for lightweight tasks. You configure which providers handle which task types. (Provider-expansion spec ratified; the routing layer is the next build.)
  • Pen & Paper RPG Design Tabletop RPG design in analog mode. Dice system design, class building, encounter balance, bestiary generation. Pulls from open SRDs (d20, Fate, Powered by the Apocalypse). Generates fillable character sheets, GM screens, and random tables.
  • Art Asset Pipeline Free, browser-based art generation. Base sprites, animated loops, sprite-sheet conversion, with ComfyUI as a local alternative.
  • Stack-Aware Test Automation TEST mode generates real test suites using the right framework for your stack. Run from a button, see pass/fail in the activity feed. Failed tests auto-route to Debug with context. Config validation for Ikemen GO characters, mod structures, ROM headers. Works across all 20 platforms. (TEST mode REPORT shipped in v0.3.1; the suite-generation half is the un-shipped part.)
  • Anonymous Usage Insights Opt-in, transparent, no personal data. You can view exactly what gets sent before opting in. Off by default.

v0.5 — Ship & Share

  • Localization & Translation Extract strings, translate to any language, generate locale files in the right format for your stack. Syncs translations when source text changes. Lives in PACKAGE mode.
  • DevForge MCP Server Use DevForge's modes, skills, and context system from inside VS Code, Cursor, or raw Claude Code. Free and open source.
  • Remote Dispatch Message DevForge from your phone via Telegram. Send tasks to a terminal tab, ask the ASK bot what Claude is doing. The bot confirms before expensive tasks and respects your daily budget.

Idea Bank

  • Sprite Visualizer Watch pixel characters act out what Claude is doing. Reading a file opens a book. Running tests pulls a lever. Activity feed drives the animation.
  • Companion App Native iOS/Android app for remote dispatch, ASK bot, and usage monitoring.
  • Voice Input Speak prompts instead of typing. Mic button next to BUILD & SEND.
  • Can It Run Doom? Yes.

Get DevForge

Buy on Itch.io // $9

Pay once, keep it. Claude Code subscription sold separately.

Not ready yet?

Drop your email. Updates only, no spam.