Voice-Activated Spells and Abilities: Cast Magic by Speaking

Shouting "fireball!" is one of the oldest fantasies in games, and most games still reduce it to a number key and a cooldown bar. With Neocortex Voice Commands, the spoken word is the cast: the player names the spell and the target, the game hears it on the player's own PC, and the ability fires.

What it looks like

Saltglass is an original co-op dungeon game set in the drowned vaults under the city of Corran. The player is the party's mage. Holding the right mouse button raises the staff and opens the voice channel; letting go releases the spell. That hold is the casting gesture, the same way a wand flick would be in a film.

The next chamber holds a skeletal archer on a broken balcony and two shieldbearers wading toward the party through knee-deep water. The player raises the staff:

"Fireball on the archer."

They let go, and fire arcs up to the balcony. The shieldbearers close in, so the next cast is a single word: "Shield." A ward snaps up just as the first blow lands. On the balcony, the archer pulls itself back together out of the rubble. "Cast it again." The fireball goes back up to the balcony, because "it" carries over from the last command.

Then one sentence for two spells: "Freeze the water under the shieldbearers and blink to the stairs." Ice locks both enemies in place and the mage is already on the steps behind them.

Chained to the mage's belt is Quillon, a grimoire with opinions and an AI character. Short casts go to the on-device engine; a real question goes to him:

Player: Quillon, what are these things weak to? Quillon: Drowned dead hate heat, apprentice. You'd know that if you'd read chapter four instead of sleeping on it.

Quillon knows his monsters because the game's bestiary is uploaded as his knowledge base. His reply also carried an action the game plays as a page flipping open to the right entry.

How Neocortex makes it work

• On-device and instant. Recognition and matching run on the player's PC, so the spell reaches your game with no server round trip. In a fight that is the whole difference between a cast and a missed window. It works offline, too.
• Spells as actions, targets as objects. Actions are your abilities ("fireball", "frost", "shield", "blink"), objects are what they hit or where they go ("the archer", "the water", "the stairs"). Your game resolves each object to the actual enemy or spot in the room.
• Natural phrasing. "Fireball the archer", "burn the archer" and "throw fire at the archer" can all trigger the same cast. Nobody has to learn an exact incantation unless your design wants them to.
• Two spells in one breath. A single sentence arrives as separate commands, so a freeze followed by a blink plays as a combo.
• Context. "Cast it again" or "hit him again" picks up the spell or target from the previous command.
• Push-to-talk or always listening. Push-to-talk works as a casting gesture, with a charge-up while the button is held. Always listening suits a shout system where the voice alone is the trigger.
• Command sets designed in the Neocortex dashboard as actors (the player, a familiar, a summoned creature), actions and objects, each with a few example phrases.
• Multiple languages. One command set per language, so players can cast in the language they speak.

Quillon is a Neocortex AI character built in the visual node editor: a personality, a streamed voice, an emotion tag on every line, actions he can return, and a knowledge base. Knowledge base uploads are 1 on Free, 5 on Starter, 20 on Pro and 100 on Team. Send events such as "the player was hit by frost twice" and his snark follows the fight.

Ideas to build

A wizard duel. Two mages in a ring of standing stones: the player against a rival who is a Neocortex AI character. The player casts by voice during the round, and between rounds the rival taunts back in its own voice, reacting to events like "the player countered three fireballs in a row". Keep the match's sessionId and the rival can bring up round one in round three, which stings in exactly the right way.

An incantation puzzle game. Each chamber is solved by saying the right sequence of effects: "light the left brazier, then freeze the fountain". Because phrasing is flexible, the puzzle is about working out the solution, not about guessing the exact words the designer had in mind.

A monster-taming game. The player stands at the edge of the arena and calls moves like a trainer: "Ember, tail sweep on the big one. Moss, guard." Creatures are actors, moves are actions, and quantities handle the group calls: "both of you, dodge". Calling moves out loud makes the player part of the fight instead of a menu above it.

A shout system. In an open-world action game, a handful of abilities are short shouts: "push", "slow time", "call the storm". With always listening on, the player's own voice is the button, and a quiet "push" in a tense moment feels very different from a key press.

Building it

Neocortex Voice Commands runs inside PC games on the player's machine. Each matched command goes to the ability system you already have, which still owns mana, cooldowns and range.

1. List your spells. Write down every ability, who can cast it (the player, a familiar, a summoned creature) and what it can target.
2. Design a command set in the Neocortex dashboard with casters as actors, spells as actions and enemies and places as objects. Add the phrasings players will reach for, nicknames included: "fire", "fireball" and "burn" for the same spell.
3. Add Neocortex Voice Commands to your game and connect each matched command to your ability system. Your code checks mana, cooldowns and range before anything fires.
4. Make push-to-talk the casting gesture. Decide whether the spell goes off on release, and give the player a visible charge-up while the button is held.
5. Add a mentor or rival if it suits your game: a Neocortex AI character with a personality, a voice and a knowledge base of your lore, through the Web API or an engine SDK.
6. Add a command set per language your players speak.

When the player lets go and no spell matched, play a fizzle: a puff of smoke, a sputtering sound. Players forgive a fizzle. They don't forgive a staff that did nothing at all.

Voice Commands is a licensed SDK: contact us for a Voice Commands license. The AI characters that talk back, like Quillon, come with every plan, the free one included.

Voice spellcasting is one of our AI NPC, ambient AI and voice command use cases. For larger-scale orders read voice commands for strategy games, and for a party that fights beside the mage see AI party companions for games. The voice-controlled gaming page has the overview, and knowledge bases covers what a character like Quillon can know. Start with one spell and one target, and add the rest of the spellbook once it feels good to say.

Checked October 2026.

Frequently Asked Questions