Unity ML-Agents alternatives

Open-source starter kit for a town of chatting AI characters.

• Best for: Developers who want working code for a simulation where many LLM-driven characters move, talk and remember
• Pricing: Free and open source (MIT); you pay for any hosting and cloud model use, or run models locally with Ollama
• Export: You own the code and can deploy it to Convex cloud, a self-hosted Convex backend in Docker, or Fly.io
• Watch out for: Built around the Convex backend; the README says the simulation pauses after 5 minutes of idle browser time and that changing character data means wiping and reloading the database

Talking AI characters for Unity, Unreal and the web.

• Best for: Developers who want ready-made voice NPCs with memory, actions and lip sync in Unity or Unreal
• Pricing: Free plan (100 interactions a month); Indie Dev $29/month ($22/month billed yearly); higher tiers up to $1,199/month (as of October 2026)
• Export: Not applicable: characters run on Convai's cloud and are reached from your game through its SDKs and an API key
• Watch out for: Convai's Unreal plugin builds only for Win64 and Android targets, according to its compatibility docs, and characters need a network connection to Convai's servers.

Voice, speech and LLM APIs for AI characters.

• Best for: Teams building cloud-powered talking characters who want text-to-speech, speech recognition and LLMs from one provider
• Pricing: Pay as you go (On-Demand, $0/month) with TTS-2 at $25 per 1M characters and speech recognition at $0.15/hour; monthly plans from $25 lower the rates (as of October 2026)
• Export: Not applicable: Inworld is a usage-billed cloud service, so a game that uses it calls Inworld's APIs while it runs
• Watch out for: Inworld retired its no-code Character Studio, so characters are now built in code with its Agent Runtime SDK templates, according to its deprecation notice.

Run local LLM characters inside Unity.

• Best for: Unity developers who want AI characters that run offline on the player's device at no per-use cost
• Pricing: Free and open source (Apache-2.0) for personal and commercial use
• Export: Models and code ship inside your own Unity build; no service account is needed
• Watch out for: On mobile, the README says models of up to 1 to 2 billion parameters are the practical limit, and its default model downloads run to gigabytes.

Local LLMs for Godot NPCs and other platforms.

• Best for: Godot developers who want offline AI characters with tool calling and speech, at no per-use cost
• Pricing: Free and open source (EUPL-1.2); the README says it can be used in proprietary and commercial projects
• Export: Runs inside your own build; the Godot binding exports to desktop and Android
• Watch out for: The Godot binding has no iOS or web export, and Windows ARM64 isn't supported yet, according to the README.

NVIDIA models and SDKs for on-device game characters.

• Best for: Studios building voice-driven characters that run locally on players' GPUs, in Unreal or a custom engine
• Pricing: Not published; Unreal plugins and the Audio2Face-3D SDK are MIT-licensed, the Game Agent SDK is Apache-2.0, and models use a mix of open and NVIDIA licenses
• Export: Models and plugins run inside your own game build; check each model's license before shipping
• Watch out for: NVIDIA's ACE page lists Unreal Engine 5 plugins (5.4 to 5.7, varying by plugin) and C/C++ SDKs, but no Unity plugin.

Free AI NPC APIs served through a player app.

• Best for: Indie developers and modders who want AI NPCs without paying for model APIs
• Pricing: Free for developers and players; Player2 says it pays the API costs, with free models plus daily credits for frontier models
• Export: Your game ships normally, but its AI features rely on players running the Player2 app
• Watch out for: Players need to install and run the Player2 app, which serves the AI APIs on their machine, according to Player2's Godot plugin announcement.