OpenAI Unity

Neocortex's pitch is honest about where the work usually is: not in getting an LLM to produce text, but in making a character that remembers, reacts to the scene, and triggers real game behavior. The node-based editor is the centerpiece — it gives you a bird's-eye view of a character's logic in one place, with conversation flows, environment interactions, and rules defined visually. That is a different shape of tool from a chat API. The feature set backs it up. Knowledge Bases supply the memory and world context most game NPCs lack; Spatial Awareness goes past static editor data to pick up instantaneous changes and objects in the scene; and the Action System is what separates a character from a chat window, firing custom behaviors rather than returning a paragraph. Emotional Intellect ties response sentiment to avatar facial-expression software, so a character speaks and visibly emotes — and because any 3D avatar works via the SDK, you're not locked to a vendor-supplied model. Multi-language support covers global releases. For education and training, the Team tier's Quiz Evaluation Engine & API (20 question sets) is a genuinely different capability rather than a bigger quota. Voice is treated as a first-class path: voice-controlled interactions and voice-aware dialogue, with custom voice creation unlocking at Pro (5) and Team (20). The constraint is the credit model. Free gives you 2 characters, 2 conversation flows, and 200 monthly credits — enough to prove the concept, not to launch. Starter at €9/mo (billed €108 yearly) raises it to 5 characters and 2,000 credits but is capped at personal and academic use. Commercial use starts at Pro (€29/mo billed €348 yearly), where you get 20 characters, 10,000 credits, Multi-NPC Group Chat, and 5 custom voices. Team at €99/mo (billed €1188 yearly) is the collaboration tier: 5 seats, 100 characters, 40,000 credits, the Quiz Evaluation Engine, and 20 custom voices. Every paid tier is billed yearly, so the €9/€29/€99 headline is the annual-equivalent rate, not a monthly invoice. Two other things to weigh before committing. The platform abstracts the underlying models, so if your project depends on choosing or fine-tuning a specific model, this is the wrong layer. And self-hosted or private-infrastructure deployment sits on the Custom plan, which is contact-sales. Against Inworld AI, Neocortex trades that model control for a visual editor and flatter pricing — a good trade for a studio whose bottleneck is engineering time, a poor one for a team whose bottleneck is model behavior. The June 2026 acceptance into the Cloudflare for Startups program is a reasonable signal about infrastructure investment, though it's early. Best fit: indie devs, VR/AR teams, edtech projects, and interactive storytellers who want characters that remember and react without hiring an ML engineer.