I’m building Vifu, an open-source AI runtime and web creator for making AI-native games. Vifu is designed to work with any game rather than being tied to one engine. It can run through a cloud endpoint, locally alongside the game, or eventually be compiled for the target device.
What It Is
Vifu decouples AI state and orchestration from traditional game engines, it builds with single Rust Binary, simple self-Hosted stack with Web Creator Extension.
Why I Built This
I ran into several specific brick walls when trying to integrate AI Agents into games:
1. AI NPC debugging in game engines is broken: Tools like Langfuse exist, but they aren't designed for game loops and require complex observability setups. Trying to debug prompt chains, context, and state inside Godot, Unreal, or Unity editor loops directly is painful.
2. Fixed provider lock-in: Platforms like Convai or UEPN lock you into their predefined provider pipelines.
3. Lack of local provider support: Existing solutions rarely support self-hosted setups. I wanted to use local providers like self-hosted Whisper or OpenClaw to build unique, custom agent and runtime combinations.
4. On-device AI on hardware: Scenarios requiring on-device AI for IoT and hardware are growing. A cross-compilable Rust runtime allows you to run locally on target hardware or delegate to a cloud endpoint.
Why I Started with Interactive Short Dramas
I initially spent a massive amount of tokens trying to build a full "Game Generation Agent." I quickly realized that maintaining code quality for raw LLM-generated game code is nearly impossible.
I pivoted to a different workflow: build the creator first, then build games on top of that editor using tools like Codex or Claude Code.
To test this flow, I built an interactive short drama / AI NPC roleplay game. Because the structured editor made NPC logic and state drastically easier to debug, I was able to build the editor, iterate on it, and ship the entire short drama game in a single day.
I'd love to hear your thoughts on this architecture and how you currently manage AI resources in your game stacks.
I’m building Vifu, an open-source AI runtime and web creator for making AI-native games. Vifu is designed to work with any game rather than being tied to one engine. It can run through a cloud endpoint, locally alongside the game, or eventually be compiled for the target device. What It Is
Vifu decouples AI state and orchestration from traditional game engines, it builds with single Rust Binary, simple self-Hosted stack with Web Creator Extension.
Why I Built This I ran into several specific brick walls when trying to integrate AI Agents into games:
1. AI NPC debugging in game engines is broken: Tools like Langfuse exist, but they aren't designed for game loops and require complex observability setups. Trying to debug prompt chains, context, and state inside Godot, Unreal, or Unity editor loops directly is painful.
2. Fixed provider lock-in: Platforms like Convai or UEPN lock you into their predefined provider pipelines.
3. Lack of local provider support: Existing solutions rarely support self-hosted setups. I wanted to use local providers like self-hosted Whisper or OpenClaw to build unique, custom agent and runtime combinations.
4. On-device AI on hardware: Scenarios requiring on-device AI for IoT and hardware are growing. A cross-compilable Rust runtime allows you to run locally on target hardware or delegate to a cloud endpoint.
Why I Started with Interactive Short Dramas I initially spent a massive amount of tokens trying to build a full "Game Generation Agent." I quickly realized that maintaining code quality for raw LLM-generated game code is nearly impossible.
I pivoted to a different workflow: build the creator first, then build games on top of that editor using tools like Codex or Claude Code.
To test this flow, I built an interactive short drama / AI NPC roleplay game. Because the structured editor made NPC logic and state drastically easier to debug, I was able to build the editor, iterate on it, and ship the entire short drama game in a single day.
I'd love to hear your thoughts on this architecture and how you currently manage AI resources in your game stacks.
The docs: https://vifu.dev