ElevenLabs Agents for Game NPCs - A Unity-Ready Alternative

Can ElevenLabs Agents power the NPCs in your Unity game? It can hold a spoken conversation, but the Unity client, the link to your game state and the cost controls are left for you to build.

ElevenLabs Agents (marketed as ElevenAgents, and formerly called Conversational AI) wraps speech recognition, a language model and text-to-speech into one managed voice pipeline. ElevenLabs also markets its voices to game studios, so it is a natural first stop when you want an NPC that talks. This post covers what the platform gives a game team, what is missing for a Unity project, and how Neocortex covers the same ground with characters built for games.

What is ElevenLabs Agents?

It is a platform for voice agents that hold conversations on websites, in mobile apps and over phone lines. Out of the box you get:

• A managed voice pipeline with turn-taking, interruptions and timeouts.
• A visual workflow builder for multi-step conversations.
• A knowledge base with document upload and retrieval.
• Tools that call client-side functions or external APIs.
• A choice of language models, or a custom one you connect.
• Telephony through SIP trunking and a Twilio integration, plus testing, evaluations and analytics.

Everything on that list points at customer-facing agents: support lines, booking assistants, a voice widget on a product page. None of it is wrong for a game. None of it knows what a scene, an animation state or a quest step is, either.

Does ElevenLabs Agents work in Unity?

Yes, with caveats. ElevenLabs publishes an early-stage ElevenAgents SDK for Unity. It targets Unity 6.3 LTS and later, runs in the Editor, on standalone, mobile and WebGL, and carries a warning that its APIs may change without notice. It connects a scene object to a voice agent: walk up, the agent greets you, walk away and the session ends. Its other official agent SDKs cover JavaScript, React, React Native, Python, Swift, Kotlin and Flutter.

What the Unity SDK does not give you is the game side. A Unity team still builds:

1. Gameplay actions: turning what the agent says into door opens, item hand-overs and quest steps, with the right target in the scene.
2. Scene awareness: telling the agent which objects are nearby and what just happened, every turn.
3. Emotion and animation: mapping replies to facial expressions and animation states.
4. Story structure: keeping a free conversation on track through quest beats.
5. The same work again for every other engine you target, and a plan for an SDK that may change under you.

With Neocortex, that layer already exists. Install the Unity SDK (Unity 6, six sample scenes), open a voice sample, press Play and talk to a character. The SDK handles microphone input in WebGL builds too, and the same characters work from Unreal Engine 5, Godot (beta) or the Web API in any language.

What does per-minute pricing mean for a game?

ElevenLabs lists its agent pipeline at $0.08 per minute, with a higher burst rate of $0.16 per minute, and passes language model costs through on top. Plans include a block of minutes and cap concurrent conversations, from 4 on the entry tier to 40 on the highest tier listed.

Run the numbers for a narrative game. If 2,000 players each spend 15 minutes talking to NPCs in launch week, that is 30,000 minutes, or $2,400 at the base rate before any model costs. Players who love your characters talk more, and a per-minute bill grows with exactly the engagement you want. Concurrency matters too: on launch night, the number of players speaking at the same moment is what hits the cap.

Neocortex uses monthly plans with included credits instead:

Testing a character in the browser editor never uses credits, so iterating on dialogue costs nothing. For a wider look at billing models, see our AI NPC pricing comparison.

What makes a voice agent a game character?

A voice agent answers questions. A game character has to stay in a role, know its world and change something in the game when the player asks.

In Neocortex you build that in a visual node editor rather than a long prompt. Agent, Personality, Emotion and Knowledge nodes define who the character is, with PDF or TXT files as its knowledge base. World, Objects, Intents and Actions connect it to your game, and Language, Accent and Voice nodes decide how it sounds.

At runtime your game sends nearby objects and recent events, and replies come back as lines with an emotion tag plus actions such as { "name": "OPEN_DOOR", "targetId": "door_02" }. Conversation Flows add branching story beats, and the API returns checkpoint keywords (flowState) your game reacts to. On Pro and Team plans, up to six characters can share one conversation over the API.

For the voice itself, clone one from a short sample you have the rights to, or design one from a text description.

How do ElevenLabs Agents and Neocortex compare?

Start with one character

Pick the NPC your players will talk to first: a shopkeeper, a guide, a rival. Build it in the Neocortex editor, give it a voice, and talk to it in the browser until it sounds right. Then drop it into a Unity sample scene and hear it answer from inside your game.

Competitor details were last checked in October 2026.

Frequently Asked Questions