X&Immersion Alternative for On-Premise AI Characters

A studio is building a training simulator for a site with no internet line. The client wants a talking instructor who answers questions about safety procedures, speaks every line aloud and reacts to what the trainee does. All of it has to run on hardware inside the building.

Teams in that position often find X&Immersion, a Paris company that builds local AI tools for games. If you are looking for an X&Immersion alternative, the real decision is between separate tools for dialogue, voice and faces, and one character platform that runs on your own hardware with a team behind the deployment.

What X&Immersion ships

X&Immersion describes itself as building infrastructure for local AI in games. Its models are tuned for size and memory so they can run at game runtime, deployed on private servers or on the player's device. The main tools:

• DiaGen generates in-game text: dialogue, quests, item descriptions and tutorials. Its runtime version writes NPC dialogue during play, with dialogue memory, retrieval from game documents and text-to-speech with custom voices.
• Ariel generates voices, for production voice-over or in real time for live NPCs, through an API call or local inference on the player's machine. Ariel Create is a separate tool for prototyping temp lines with your own voice.
• Geppetto generates lip-sync and facial animation from audio, as a production tool and in real time. It is listed on the Unity Asset Store and on Fab.
• Other tools include an in-game help chatbot, a player log summarizer and private AI copilots for studio teams built on open-source models.

The site lists integrations with Unity, Unreal Engine and Wwise, and offers custom work for other engines. Pricing is not published: each tool starts with a demo request, and the company also takes on R&D projects for individual studios. Its site links games such as Amerzone: The Explorer's Legacy and Hot Lap Racing among its projects.

Does X&Immersion run on the player's device?

Yes, for its runtime tools. DiaGen, Ariel and Geppetto each have a real-time mode that can run locally, and Ariel and Geppetto can also run through an API call.

Running on the player's device shapes the whole project:

• Hardware floor. The model shares the GPU and memory with your game, so every target machine needs headroom for both.
• Smaller models. A model that fits next to a game is smaller than one on a server, which limits how long a character holds its persona and how much it knows.
• Build size. Model files ship in the build or download on first launch.
• Per-device testing. Each GPU tier, console or phone you target needs its own performance pass.
• Joining the tools. Dialogue, voice and lip-sync are separate products, so your team connects them and decides how each one talks to your game logic.

These costs come with any on-device approach, whichever vendor you pick. Our comparison of AI NPC tools with local AI support and our look at Tryll Engine cover the same trade-offs from other angles.

X&Immersion alternative: tools or a character platform

There are two ways to build local AI characters.

Assemble tools. You pick a dialogue model, a voice model and a facial animation system, then wire them into your engine. This suits studios with engineers who want control of each layer and already run a character pipeline of their own. The trade-off is ownership: your team maintains the joins, and every change to a personality or a fact goes through whoever built the pipeline.

Use a character platform. Personality, knowledge, voice and game actions live in one character definition that writers can edit, and the engine integration is one SDK. The trade-off is that you work within the platform's model of a character. For NPCs, companions, guides and instructors, that is usually the model you would have built anyway.

Neocortex is the second kind. Characters are built in a visual node editor with Agent, Personality, Emotion, Knowledge, World, Objects, Intents, Actions, Language, Accent and Voice nodes. Knowledge takes PDF or TXT files. Conversation Flows split a scene into story beats and return checkpoint keywords (flowState) your game reacts to. Each reply carries a line, an emotion tag and actions such as { "name": "OPEN_DOOR", "targetId": "door_02" }. Facial animation stays in your own pipeline, and the emotion tag gives it a cue for every line.

How Neocortex runs on your own hardware

The standard Neocortex product is self-serve and runs in our cloud, built for real-time conversation. For studios that need AI characters on-premise, offline or on-device, we offer a Custom (enterprise) arrangement, set up through a conversation with our team:

• The full platform on your servers or private cloud. The character editor, the API, your characters and Conversation Flows run inside your own infrastructure.
• Local voice. Speech-to-text and text-to-speech run on-premise too, not only the language model.
• Air-gapped operation. Neocortex can run with no internet connection at all, for secure sites, training simulators, kiosks and events.
• On-device inference. Inference can ship inside the game build and run on the player's machine.
• Support from our team. We work with your studio on deployment, model choice and integration.

Measured speed. On a laptop RTX 3080 (8 GB), our on-device runtime returns its first token in 0.06 seconds and starts speaking in about one second, because it voices the first sentence while the rest is still being generated. Speech recognition and text-to-speech run on the CPU, so a 3B model needs about 3 GB of VRAM, and a 2B model generates over 300 tokens per second in under 1 GB.

On-premise deployments are sized with each studio, around the number of conversations you expect, the hardware you have and the model you choose. If that describes your project, contact us and tell us where your characters need to run.

Everything else is the product self-serve users already get: SDKs for Unity (Unity 6, six sample scenes), Unreal Engine 5 and Godot (beta), a Web API for any language, voice cloning from a short sample you have the rights to, voice design from a text description, and group chat with up to six characters in one conversation on Pro and Team.

Who this route is for

Neocortex on-premise fits studios that:

• build training, defence, healthcare or education projects where voice and text cannot reach a third-party cloud,
• ship to museums, venues, kiosks or events with unreliable or no internet,
• have a client or publisher contract that requires the backend to run in their own infrastructure,
• want writers to edit characters in an editor while the whole stack stays inside the building.

If your game is online anyway and your players run a wide range of hardware, the cloud product is the simpler start, with plans from free to Team. Our guide to AI NPC pricing models covers the cost side.

Example: an offline museum guide

Picture a museum installation: a historical figure on a kiosk who answers visitors' questions out loud and points them to the next room.

With separate local tools, the studio picks a dialogue model and feeds it the exhibition texts, adds a voice model, and writes the code that passes text from one to the other and decides when to send visitors onward.

With Neocortex on-premise, a writer builds the guide in the editor: Personality for the voice of the period, Knowledge with the exhibition catalogue as a PDF, Intents for common questions, and an Action such as SHOW_ROOM that the kiosk app reads to highlight a room on its map. The platform, speech recognition and voice run on a machine in the museum with no internet line. When the exhibition changes, the team updates the Knowledge files in the editor and the kiosk app stays as it is.

The bottom line

X&Immersion sells local AI as a set of tools for dialogue, voice and faces, plus R&D work for studios. Neocortex gives you the character itself: designed in an editor, voiced, wired to your game through one SDK, and available on your own servers, fully offline or inside your build, with our team alongside you. Build a character in the cloud editor for free to see how it behaves, then talk to us about running it where your project needs it. For the wider field, read our roundup of local and on-premise AI NPC platforms or browse all Neocortex comparisons.

Competitor details were last checked in October 2026.

Frequently Asked Questions