CLOSED TECHNICAL PREVIEW — NOT PRODUCTION READY

What Happens to Your Frame Rate
If You Double the Population Right Now?

If the answer is "it crashes," how do you plan to ship? Behavior trees scale linearly with complexity. Your CPU budget doesn't. That's not a coding problem. That's a math problem. You will lose.

The 6-Phase Epoch

Every tick, every agent goes through the same pipeline: Perceive → Strategize → Think → Collect → Apply → Reconcile. This isn't a suggestion. It's enforced. No spaghetti logic. No race conditions. No "why did this agent do that?" mysteries.

Agents use utility-based BDI for crowds (Tier 1/2) and GOAP planning for complex NPCs (Tier 3). They don't follow scripts. They have beliefs, desires, and they plan actions to satisfy them. Emergent behavior, not scripted responses.

// CORE_CAPABILITIES
BDI/GOAP Runtime FULL SUPPORT
Deterministic Execution BIT-EXACT REPLAY
Signal-Based Architecture EVENT-DRIVEN
Utility-Based Decision Making SCORE-DRIVEN
Forensic Debugging REASONING TRACES
DOTS/ECS Integration NATIVE SUPPORT

The Replayability Value Proposition

I'm not selling you a library. I'm selling you the ability to replay a bug from a user's save file instantly. How many hours of QA time does that save you in a year? That's the value. Bit-exact determinism means you can take their save seed, put it on your machine, and it crashes exactly the same way at exactly the same tick. Debug the past. Don't guess the future.

HOW WE GUARANTEE IT:

Q16.16 Fixed-Point Math — No floating-point drift. Quantized before arithmetic, not after.
PCG32 Deterministic RNG — Seed-based randomness. Same seed = same sequence, every time.
Canonical Iteration Order — Agents processed in sorted ID order. No platform-dependent iteration.
Per-Tick State Hashing — xxHash64 verification at every tick. If it drifts, you know immediately.

If it crashes on their machine, you take their save seed, put it on your machine, and it crashes exactly the same way at exactly the same tick. Debug the past. Don't guess the future.

// DETERMINISM_GUARANTEES
Cross-Platform Replayability BIT-EXACT
Fixed-Point Math QUANTIZED FLOATS
Deterministic RNG SEED-BASED
No Floating-Point Drift VERIFIED

The Performance Budget

You can try to optimize that behavior tree, but that's O(n²) complexity. You aren't fighting code anymore; you're fighting math. We built a kernel that guarantees 30 TPS at 10,000 agents. It's zero-allocation. It's deterministic. It stops the bleeding. But you have to rip out your current AI to use it.

WE DON'T "OPTIMIZE" LATER. WE ENFORCE IT IN THE KERNEL.

10,000 Agents? 0 Bytes allocated per tick.
Complex Planning? 0 Bytes allocated per tick.
Spatial Queries? 0 Bytes allocated per tick.

If a developer on our team allocates memory in the hot loop, the build fails. We don't trust ourselves, so you can trust the runtime.

// PERFORMANCE_GATES VERIFIED
GATE-PERF-01 [SCALE] PASS (10,000 AGENTS)
GATE-PERF-02 [TPS] PASS (>= 30.0 FIXED)
GATE-MEM-04 [ALLOC] PASS (0 BYTES HOTLOOP)
GATE-DET-01 [REPLAY] PASS (BIT-EXACT)
* Reference Hardware: Intel i7-13700HX, 32GB RAM, RTX 4060 Laptop. Performance may vary by platform.

The Separation of Church and State

Your Game Engine (Unity) is for rendering, physics, and feeling. Warweaver is the Brain. They communicate via a strict telemetry protocol. The Brain runs the simulation (BDI, GOAP, Economy). The Body visualizes the result. This isn't a plugin. It's an operating system for emergent behavior.

Supported Runtimes

Warweaver is currently focused on Unity. We're building the best product we can for Unity developers. Support for additional engines may be added in the future based on demand.

// RUNTIME_COMPATIBILITY
Runtime Version Package Status
Unity 6.0+ Unity Package Manager SUPPORTED
Additional engine support (Unreal Engine, Godot, .NET Standard) may be added in the future based on demand.

The Disqualification Test

Honestly, for a shooter, Warweaver is overkill. You should stick with Unity NavMesh. Save your $300. But if you're building Rimworld, Kenshi, Dwarf Fortress—systemic NPC at scale—and you're spending 40% of your sprint debugging AI logic, you can't afford NOT to use this.

This is a closed technical preview. Not production-ready. The $300 fee filters serious teams from hobbyists. Request evaluation access.

EVALUATE SDK