The Halloween AI-made game you can buy and play now

CODEX MORTIS is a released dark-fantasy action game that offers a more concrete test of AI-assisted game development than a speculative trailer. It began Steam Early Access on March 26, 2026, then reached version 1.0 on September 24. The developer, GROLAF, and publisher, CRUNCHFEST, say generative AI was used across the game's code, text, artwork, sound, and music. The result is a conventional, downloadable survivors-style game with unusual production methods. Its necromancers, skeleton armies, and cursed arenas make it a timely Halloween recommendation without pretending it launched this weekend.

The official Steam store page at https://store.steampowered.com/app/4084120/CODEX_MORTIS/ lists the complete game for purchase, with solo play, local cooperative play, and online cooperative play. This is not a free browser game, and the Steam page is the verified route to the product. The 1.0 announcement confirms a substantial launch rather than a renamed prototype: new magic schools, more heroes, four additional stages, and four-player online co-op. October's subsequent balance patches provide a second reason to examine it now. They also show that shipping software with AI tools does not remove the ordinary work of balancing, fixing, and supporting a game after release.

How the spell-deck roguelite actually works

The basic loop combines two familiar genres. Like a horde-survival action game, CODEX MORTIS throws growing crowds of enemies at a hero who must stay alive while attacks fill the arena. Like a deckbuilder, it asks players to choose a collection of spells before a mission. The official description says you assemble a deck from the spells you own, carry that loadout into battle, and earn points that can buy booster packs afterward. New cards expand the collection for future attempts. This makes preparation meaningful: a powerful-looking spell is less useful if the rest of the deck cannot support its timing, positioning, or resource demands.

An illustrative strategy would be to build around summoning so that undead allies occupy enemies while the player keeps moving. Another would emphasize risky Blood Magic, trading survivability for offensive power. These are examples of strategic directions described by the store, not a claim that we personally tested an optimal build. Keep an eye on which spells are consistently useful against ordinary crowds, which help with bosses, and which are only impressive when several conditions align. The point of the card system is experimentation within rules, not unlimited generation from arbitrary prompts.

Ten schools, 200 spells, and mutations

Version 1.0 expanded the game to ten schools of dark magic and a stated total of 200 spells. The store identifies Necromancy, Summoning, Blood Magic, Soul Magic, Curses, Harvest, Death Knight, Spiritism, Desolation, and Apothecary. The five additions at the September launch brought scythe attacks, close-range sword techniques, spirit turrets, hostile zones, and alchemical effects into the existing combinations. The developer says every school can interact with the others through synergies and mutations. That creates a wide design space, although the number of possible combinations is not the same as the number of distinct, well-balanced playstyles. The actual value depends on how reliably the mechanics cooperate.

Mutations change how a spell behaves rather than simply increasing its damage number. Steam gives examples involving a bone spear taking a new pattern, summons changing their undead archetype, and area effects covering different ground. A player who has learned to dodge around one projectile shape may need to relearn the timing after a mutation. That is where the game can be more interesting than a spreadsheet of numerical upgrades. A sensible test is to compare one familiar spell before and after a mutation, then ask whether the change creates a tactical decision. If every combination only makes the screen brighter, variety may be less meaningful than the headline spell count suggests.

Thirty-one heroes and a much larger endgame

The September release raised the roster to 31 playable heroes, including characters built around combinations of two or three magic schools. GROLAF's launch notes name Ravena Bloodscythe as a Blood Magic and Harvest specialist and Seraphon Wraithblade as a blend of Harvest, Death Knight, and Soul Magic. Those examples suggest that hero selection changes the starting direction of a build, rather than serving only as a cosmetic skin. The game also lists fourteen levels across its campaign and other modes. Four stages added to the Apocrypha at 1.0 include The Garden of Returned Flesh, The Golden Rot, The First Mercy, and The Road to Blood Bath.

The developer describes the final Road to Blood Bath encounter as a three-phase boss with 600,000 health. Boss design depends on attack patterns, movement demands, build variety, and how long the encounter lasts. Some Steam players have criticized lengthy high-health encounters, while others praise the spell combinations. Newcomers should expect a progression-heavy action game and should not assume that every one of its fourteen stages plays like an entirely separate genre.

Four-player online co-op is part of the finished release

Online cooperative play for up to four people was a major feature of the September 24 version 1.0 launch. The official announcement says players can host through Steam, join friends, use a password or lobby code, and rejoin an ongoing run after a disconnect. Local cooperative play also remains available. That distinction matters for groups: two people on one computer and four people connecting over the internet are different arrangements with different setup needs. The Steam listing identifies both online and shared-screen cooperative features. It is a real advertised mode in the released product, not a future stretch goal or an experimental feature mentioned only in a press interview.

Multiplayer also exposes problems that a single-player build can hide. The developer's September 28 patch notes mention an experience-orb synchronization fix for online sessions and the restoration of local co-op on Linux and Steam Deck. Those are specific examples of post-launch maintenance. If you are planning a Halloween group session, have everyone update the game, confirm that Steam's networking is available, and try a short run before committing to a long one. A host with a different version or a problematic connection can complicate the experience. We have verified that the feature is advertised and patched, but have not independently tested a four-person session or measured its network stability.

What the October 9 and 10 balance updates changed

The October 9 developer announcement is especially revealing because it acknowledges that progression felt uneven. Some players unlocked content too rapidly, others stalled early, and certain bosses became drawn-out fights. The published overhaul addressed card-pack pricing, achievement rewards, permanent upgrades, difficulty selection, boss scaling, and the pacing of Story and Survivor modes. It also separated Mortis Mastery growth from the challenge setting selected before a run. That is a conventional but important design decision: players should be able to improve their character without discovering that the game silently raised enemy difficulty at the same time. The developer framed these adjustments as responses to feedback, not as a demonstration of AI magically balancing itself.

On October 10, the team published a follow-up restoring parts of the previous progression system. It brought back the earlier Mortis Mastery bonuses, higher-rarity multi-skill level-up rewards, and soul reward multipliers across maps. The rapid follow-up suggests the first overhaul did not satisfy every expectation. It would be inaccurate to describe all October changes as permanent improvements without examining player reactions and subsequent builds. For readers returning after September, the practical advice is to review the current upgrade descriptions rather than rely on an older guide. For anyone evaluating AI-built games, these patches are a useful reminder that a working product still needs human judgment about pacing and enjoyment.

What does “100% AI-developed” actually establish?

Steam's developer disclosure states that AI was used for the entire game's code, text, art, sounds, and music. The publisher's September announcement also points to a documentary describing development from a chat-window prototype to a commercial release. These are substantial disclosures, but the phrase “100% AI-developed” should be attributed to the people making the game rather than treated as an independently audited measurement. A human still chose goals, evaluated output, directed revisions, published builds, and responded to players. AI assistance across every content category is not the same thing as an autonomous system deciding on its own to design, ship, and operate a game.

There is another important distinction: the Steam page describes generative AI used during development, not necessarily live text or art generation while someone plays. CODEX MORTIS therefore belongs in a different category from an AI-NPC conversation game, where a model is called to answer each player's dialogue. Both are AI-related games, but they raise different questions. Here the central issue is whether AI-assisted production can deliver coherent mechanics, readable art, reliable controls, and a satisfying progression loop. Players can assess the finished experience without needing to endorse every claim about its creation. Developers can learn from its rapid iteration while still asking how quality assurance, authorship, and asset provenance were handled.

Steam reviews, limitations, and what to check before buying

The current Steam listing shows a Very Positive overall review category, with the snapshot displaying roughly four-fifths favorable reviews. That is encouraging, but the sample is modest and the percentage can change. Reviewers have praised build combinations and presentation while others have complained about screen-filling effects, late-game boss durability, and progression pacing. These opinions should be read as user reports rather than universal facts. The game's many projectiles and summons can also create visibility and performance challenges, particularly when several players are casting simultaneously. Readers should inspect recent reviews and patch notes for issues that matter to their hardware and preferred mode.

Steam lists Windows and native Linux support, with Steam Deck Verified compatibility, controller support, and 28 interface languages. Its current Windows minimum requirements include a 64-bit Windows 11 system, four gigabytes of memory, and a modest older graphics card; the store lists about 140 megabytes of storage. The game is sold on Steam, and any discount displayed there is temporary and region-dependent. Do not download an unofficial installer promising a free “AI build.” If you are unsure whether the game suits you, watch current gameplay and check whether its deck preparation and boss encounters sound enjoyable.

Halloween verdict: an AI-built game worth judging as a game

CODEX MORTIS is a particularly useful Halloween 2026 example because the claim about AI production is attached to something concrete: a September 24 full release, ten schools of magic, 200 spells, 31 heroes, fourteen levels, and advertised four-player online co-op. Its October balance changes provide evidence that the developer continues to adjust the experience after launch. The supernatural setting fits the season, but this article does not claim a special October 31 event, a free giveaway, or a new Halloween patch. The real story is that an unusually AI-heavy development pipeline produced a commercially available game that can be evaluated on its own merits.

Start with the official Steam listing at https://store.steampowered.com/app/4084120/CODEX_MORTIS/ . For developer-authored details, consult the September 24 launch announcement through https://steamcommunity.com/app/4084120/allnews/ and the October 9 and 10 update notes in the same news archive. Those sources support the release date, gameplay systems, AI disclosure, and progression changes. They do not independently verify the percentage of work done by AI or guarantee flawless multiplayer and balancing. If you try it, judge whether the deckbuilding choices matter, whether mutations create meaningful decisions, and whether the battles stay readable. That is a stronger test of AI-assisted game development than either dismissing or praising a project based on its production label alone.