DEV Community

dev game
dev game

Posted on

Why Hades Bet on MonoGame Then The Forge: The Quiet Migration Behind a Million-Copy Roguelike

Why Hades Bet on MonoGame Then The Forge: The Quiet Migration Behind Supergiant's Roguelike — And What Hades II Proves

Twenty-five people in San Francisco shipped one of the best-reviewed roguelikes ever without Unity or Unreal. Hades is a 2020 isometric hack-and-slash roguelike where Zagreus fights out of the Underworld through Tartarus, Asphodel, Elysium and Styx, dying and returning to the House of Hades to grow stronger. It sold over one million copies and won Game of the Year at the BAFTAs, D.I.C.E. and GDC.

That is the short answer to "what engine does Hades use?" Early access ran on MonoGame. The shipped game renders on The Forge. PCGamingWiki lists Early access as MonoGame and Released Renderer as The Forge for Hades, and Hades II still lists The Forge as Renderer. The Forge's own GitHub names Supergiant Hades as its most notable example of writing a custom engine from scratch.

Most retellings stop at "indie custom engine, good performance." That misses the interesting bet: Supergiant treated MonoGame as scaffolding and The Forge as a shipping hull. My read after digging through Supergiant's pages, Steam, MonoGame docs and The Forge repo: Hades did not succeed despite changing renderers mid-project. It succeeded because a C# prototype let writers iterate fast, then a C low-level framework let a 25-person team ship everywhere without hiring a porting army.

The team and the constraint

Supergiant Games was founded in 2009 by Amir Rao and Gavin Simon from EA Los Angeles, with artist Jen Zee and composer Darren Korb joining early. The studio had only seven permanent staff in 2009, all still there in 2025, growing to just 25 people in 2025.

Their first game, Bastion, was built on Microsoft XNA — a managed C# runtime for Xbox 360 and Windows that Microsoft abandoned in 2013. That XNA root matters. MonoGame is explicitly a free, open-source C# reimplementation of the XNA 4 API, code-first with no built-in editor, for graphics, audio, input and content management. It began as an iPhone port of XNA in 2009 and continued after Microsoft killed XNA.

In other words, Supergiant did not pick a trendy engine. They stayed inside the C# family they already knew from Bastion, then Transistor and Pyre, and used MonoGame to keep moving while XNA itself died underneath them. Supergiant's own Hades page still pitches it simply as a rogue-like dungeon crawler where you defy the god of the dead — no engine bragging, just genre and myth.

By late 2026 that modesty looks strategic. Hades launched on Windows, macOS and Switch on September 17, 2020 after early access from December 2018, then PS4, PS5, Xbox One and Series X/S in August 2021 and iOS via Netflix in March 2024. Steam's Hades page frames it as Bastion's action plus Transistor's atmosphere plus Pyre's character storytelling, with thousands of Boon builds, the Mirror of Night for permanent growth, and God Mode for players who want story without punishment. Minimum specs are still modest: dual-core 2.4 GHz, 4 GB RAM, 1 GB VRAM and 15 GB storage.

What MonoGame actually did

MonoGame's job was to let a tiny team prototype a systemic roguelike without building platform code:

  • C# iteration. Designers and programmers already fluent from Bastion could write gameplay in C#, with content management for hand-painted art and Darren Korb's score, without learning a new editor paradigm.
  • XNA compatibility. Existing XNA concepts — content pipeline, sprite batches, DirectX on Windows and OpenGL elsewhere — carried over, so early access could ship on Steam in December 2018 while consoles were still unsolved.
  • No editor tax. MonoGame provides libraries but no built-in editor, so developers choose their own tools. For a studio that hand-places rooms from pools of pre-designed layouts with randomized order and enemies, that code-first bias fits better than a visual scene-graph overhaul.

That explains why early access lists MonoGame. It was the fastest way to test the core loop: enter rooms, pick one of three Boons from Zeus, Athena, Poseidon and others, stack damage and effects, die to Megaera, the Hydra, Theseus and Asterius or Hades himself, return to the House, upgrade, romance Megaera or Thanatos, try again.

Why ship on The Forge

The Forge is not a full engine. Its readme is blunt: it provides building blocks to extend engines or write custom engines from scratch, plus SDKs and platform ports. What is not there: physics, networking, sound. What is there: async resource loader, Lua scripting for tests, Ozz animation, Vectormath with NEON, GPU/CPU memory allocators, C input with touch gestures, flecs ECS, zip-capable filesystem, Dear ImGui UI, and an HLSL-superset shader translator.

It supports Windows 10/11 with DirectX 12/DXR, Steam Deck with Vulkan ray query, Android Vulkan, iOS 14.1/17.0 and macOS 11/14 on Intel and Apple silicon, Quest 2, plus Xbox One/Series, PS4/PS5 and Switch for accredited developers. PC, macOS, iOS, Android and Deck are under Apache 2.0 on GitHub; consoles need a commercial license.

For Hades that solves three shipping problems MonoGame alone does not:

  1. One renderer, eight targets. Hades now runs on Windows, macOS, Switch, Switch 2 via Hades II, PS4, PS5, Xbox One, Series X/S and iOS. The Forge abstracts DirectX 12, Vulkan and Metal behind one shading language, so Supergiant's hand-painted isometric art does not need three separate lighting paths.
  2. C control where it counts. Particle bursts, dash i-frames through thin walls, and hundreds of simultaneous Boon effects live or die on memory and draw-call discipline. The Forge's explicit allocators and resource streaming give a 25-person team the same levers AAA renderer teams use, without adopting all of Unreal.
  3. Keep gameplay, swap the hull. Physics, run logic, Boon stacking and Pact of Punishment modifiers could stay in Supergiant's own gameplay code while the renderer underneath changed. That is why early access players did not lose saves or feel when the backend moved.

Hades II proves the hull held. It entered early access on Windows in May 2024 and macOS in October 2024, then fully launched September 25, 2025 for Windows, macOS, Switch and Switch 2, with PS5 and Xbox Series X/S in April 2026. It keeps The Forge as renderer and earns 94 on Metacritic and OpenCritic. The sequel adds a second Surface path (Ephyra, Thessaly, Olympus, Summit) alongside the Underworld path (Erebus, Oceanus, Mourning Fields, Tartarus), a Magick Bar for Omega attacks, Hexes from Selene, and God Mode that starts at 20% resistance plus 2% per death to 80%. Two full campaigns on one custom stack from a 25-person studio is the strongest argument that the migration was not a one-off hack.

What the stack cost

  1. Two codebases to understand. New hires must learn Supergiant gameplay conventions plus The Forge's C99-style systems and FSL shaders, not just Unity C#. That narrows hiring versus a pure-Unity indie.
  2. No editor ecosystem. No Asset Store for quests, dialogue or save systems. Supergiant built House interactions, Codex, Keepsakes and Pact sliders themselves. Great for cohesion, slow for prototyping unrelated genres.
  3. Console licensing friction. Apache 2.0 covers PC and mobile, but PlayStation, Xbox and Switch need a separate commercial arrangement through The Forge Interactive. A Unity or Unreal subscription would bundle that differently.
  4. Documentation burden. The Forge's docs live in GitHub, wikis and conference slides, not in a single manual with thousands of tutorials. Small teams trade community answers for control.

None of those killed Hades. All of them shaped who can copy it.

My take: scaffolding vs. hull, and story as a resource

Here is where I diverge from "custom engine equals performance" summaries.

1. MonoGame was scaffolding, not a mistake. Most postmortems frame engine switches as failures. I think Supergiant used MonoGame correctly: as disposable velocity. Early access needed daily balance patches for Boons, weapons and Pact heats, not perfect barriers on PS5. C# let the writing team see a new Olympian interaction in-game the same day. You do not need a shipping renderer to learn that players love Artemis crits but ignore Poseidon knockback. Start in the language your writers can break safely, then harden the hull once the fun is proven.

2. Story is a progression resource, and custom tech protects it. Hades lets you romance NPCs, redecorate the House, unlock titles and hear thousands of voiced lines that only trigger after specific deaths. That requires deterministic hooks everywhere: every death must know who killed you, with what, while carrying which Keepsake. A generic quest plugin would fight that. Owning the save format, event bus and content loader meant narrative debt could be spent like currency — fail a run, earn a scene. That is why Hades feels generous while other roguelikes feel grindy. The engine did not just run fast. It remembered.

3. Boons are compiler flags. Each Boon is a small, testable transform: add lightning on attack, push on special, heal on cast. The Forge's ECS plus Supergiant's own stacking rules let thousands of combinations coexist without hand-authoring each build. If you squint, a Hades run is a program compiled at runtime from Olympian libraries. MonoGame's content system made those libraries easy to author; The Forge's memory control made them safe to combine at 60 fps on Switch. Indie deckbuilders should steal the metaphor: design buffs as pure functions, not as bespoke scripts.

4. Twenty-five is a feature. Supergiant's headcount cap is not lore. It forces the engine choice. A 25-person team cannot maintain a proprietary editor, a visual-scripting fork and a console porting department. They can maintain gameplay plus a leased renderer. The Forge's "no physics, no netcode, no sound" gap is actually a filter: if you do not need rollback or replication for a single-player isometric crawler, do not pay for it. Hades II's two-path structure reuses the same bosses, biomes and Hex systems rather than doubling content — scope discipline as engine strategy.

If you want a heresy for your next jam: do not ask "which engine is best?" Ask "which part is scaffolding and which part is hull?" Prototype in whatever your team already speaks. Ship on whatever lets you stay small.

What indie devs can steal this month

  • Prototype narrative hooks in week one. If death, upgrade and dialogue share one save event, every failed run can pay story. Do not bolt story on after systems lock.
  • Budget for a renderer swap. Keep platform code behind a thin abstraction — resource load, draw, input, file — so early access can use MonoGame, FNA or even Love2D while you evaluate The Forge, bgfx or SDL_GPU for 1.0.
  • Design buffs as data. One Boon, one file, one test. If two designers cannot combine their buffs without a programmer, your architecture will cap your build diversity.
  • Ship God Mode early. A 20%-plus-2% damage-resistance slider costs almost nothing and unlocks press, accessibility testers and story-first players who will market your game for you.
  • Stay small on purpose. If your team is under 30, prefer frameworks that do less. Every missing subsystem you do not need is a hiring plan you do not have to fund.

Train the Hades muscles in a browser

Hades demands three separate instincts — drafting under pressure, loot triage, and build commitment. You cannot practice Styx without owning it, but you can drill each muscle in short browser loops:

  • For Boon drafting and soul-stakes decisions, play Dungeon Crawl Theater. It is a roguelike deck builder where you fight for your soul, battle possessed miniatures and draft cards to embolden your deck — the same pick-one-of-three tension as choosing between Zeus chain-lightning and Athena deflect, plus the same spend-souls-to-improve-odds economy as the Mirror of Night.
  • For hack-and-slash positioning and loot greed, load Crypt Crawler. You dive into epic battles, crush monsters and collect legendary loot while customizing gear and tackling quests — a low-stakes way to practice when to dive for that extra Chaos gate versus when to surface with what you have, just like deciding to push Elysium for a Duo Boon or bank your Darkness.
  • For build-around-one-idea runs, try Dwarves: Glory, Death, and Loot. You command an auto-battling army, arrange tanks, berserkers, rangers and runesmiths, then chase fire, poison or armor-crush synergies across runs. That pre-battle composition versus mid-battle ultimate timing maps neatly to Hades weapon aspects plus Pact heats: pick a core, then bend every Boon, Hammer and Keepsake around it.

None of these are Hades. That is the point. They isolate drafting, greed and commitment so your hands learn them before Chronos tests all three at once in Hades II.

Bottom line

Hades' niche-plus-hot combo is why this story still ranks: a niche path (XNA-born MonoGame C# for early access, then Confetti's C-based The Forge for shipping) behind a now-massive roguelike (1M+ copies, 93-94 Metacritic, GOTY sweep, Hades II full release September 25, 2025). Search data backs the curiosity — Google suggests "hades what engine," "what engine does hades use" and "hades 2 what engine" as live completions, while Bing's organic route currently drops those qualifiers to dictionary noise, which is why this piece leans on primary pages instead of SERP counting.

If you remember one thing: Supergiant did not choose between iteration speed and shipping breadth. They sequenced them. MonoGame bought the fun. The Forge bought the ports. For a 25-person team aiming at story-rich systems, that was not indecision. It was the design.

Top comments (0)