AI Systems

One AI layer, orchestrated on-device

Every AI subsystem shares the same runtime core — no cloud round-trips, no glue code.

BEHAVIOR

NPC Behavior

Behavior tree system with composable nodes, state machines, utility scoring, and runtime debugging. Genre-aware presets for combat AI, patrol routines, trading NPCs, and more.

Key benefit: Ship believable NPC behavior in hours, not weeks.

VOICE

Voice Commands

On-device speech recognition and synthesis with zero cloud dependency. Low-latency voice interaction for NPCs, UI navigation, and in-game commands. Privacy-first — audio never leaves the device.

Key benefit: Voice-driven gameplay without cloud costs, latency, or privacy concerns.

NPC

NPC Intelligence

On-device small language model powering NPC memory, personality, and contextual dialogue. NPCs remember past interactions, adapt tone to player behavior, and generate unique responses — all running locally.

Key benefit: Every player gets a unique story. No two playthroughs are the same.

How orchestration & permissions work

Swipe through the controls that keep an agent fleet governed — deny-by-default, scoped, and fully audited.

Agent fleet orchestration

One runtime core coordinates every agent — planners, tool-users, and voice agents share the same scene graph and policy, with no glue code between them.

Deny-by-default

Nothing runs unless it is explicitly allowed. Every tool call and capability is closed until a policy grants it — the safe default, not an afterthought.

Role-based access control

Permissions are granted by role and scope, not per-request guesswork. Map each agent or operator to exactly the capabilities their role needs.

Immutable audit log

Every allow and deny decision is recorded. Trace exactly which agent did what, when, and under whose authority — for compliance and incident review.

For AI labs: govern your model's tool access

The same orchestration layer is what a lab inherits when its model drives RakuAI over MCP. Connect once; the permission posture below is the default — no per-vendor work.

Deny-by-default in production The 5 read tools (get_scene_state, capture_scene, get_metrics, xr_get_session, xr_list_anchors) are live. The 12 mutation tools run sandboxed until a connection is granted higher tier on the sandbox → dev → production ladder.

One versioned contract 17 native tools, drift-protected in CI — identical for Claude, ChatGPT, Gemini, Copilot. A picker showing all 147 registry entries (17 native + 7 passthrough + 123 catalog) is expected; pin to the 17.

Every mutation audited Each allow/deny decision is written to an immutable audit log — trace which model did what, when, under whose authority, for incident review.

Honest status The contract is identical by construction; per-vendor end-to-end parity is a Phase 2 deliverable, not yet proven. No lab is bundled today — this is the path we run with the first one.

Onboard your model → MCP Server

Smarter agents, games, and assistants

AI is not a plugin. It is the foundation — navigation, behavior, voice, NPC intelligence, and a governed MCP surface any model drives from day one.

For AI Labs Explore the SDK Spatial Engine

Microsoft for Startups

RakuAI is an NVIDIA Inception member. We are bringing NVIDIA-accelerated 3D Gaussian splatting to Raku Capture — a phone scan of any room turns into an explorable splat in minutes. Read more →

NVIDIA Inception is a free program for AI startups; RakuAI's membership does not imply that NVIDIA endorses RakuAI's products. Learn more at nvidia.com/startups.

RakuAI is built on AWS and an AWS Activate participant. Production reconstruction runs on Azure today; AWS GPU capacity (G5 / G6e) is on our roadmap, honestly gated until quota lands.

“AWS” and “AWS Activate” are trademarks of Amazon.com, Inc. or its affiliates. RakuAI’s participation does not imply that AWS endorses RakuAI’s products.