Skip to content

yuechen-li-dev/Dominatus

Dominatus

NuGet NuGet Downloads License: MIT

Dominatus is a high-performance, deterministic .NET agent runtime kernel designed for complex automation, simulations, and AI agent orchestration.

The industry is learning the hard way that LLMs are terrible, expensive schedulers for high-frequency workflows. Dominatus takes agent orchestration out of the "haunted prompt-chain loop" and brings it down to the native CPU layer.

By executing agent behaviors as highly optimized, local C# control flow - using Hierarchical Finite State Machines (HFSMs), Utility AI, structured blackboards, and policy gates - Dominatus provides a bulletproof, zero-allocation control plane. LLMs are treated as optional yet powerful semantic reasoning actuators on the boundary, not the main scheduler.


⚡ Hard Benchmark Receipts

Dominatus drops agent execution costs and latency to absolute zero by running ticks locally instead of making blocking, expensive model calls over the network.

The following metrics are tracked in our headless CPU simulation workload (net10.0 Release, Ubuntu 24.04, 2 Cores):

Metric Performance What it Means for Architecture
RTS Skirmish Simulation 130,862 agent-ticks/sec Ultra-dense behavioral simulation loops at true CPU speed.
Utility Option Evaluations 1,177,761 evaluations/sec Lightning-fast Ai.Decide scoring across massive option matrices.
State Replay & Resume Deterministic Hash Match 100% deterministic checkpoint/resume. Perfect for state audits and bug replication.
Parallel Decision Modes Identical Hash Equivalence Multi-threaded execution pipelines yield identical state hashes to sequential runs.

💡 The Bottom Line: Dominatus doesn't make prompt chains faster. It makes prompt chains irrelevant.


🎯 How Dominatus Fits the Ecosystem

Dominatus occupies a completely different layer than existing AI framework wrappers. It functions as the operating system kernel for state management, while other tools act as userland plugins.

System Main Role The Dominatus Difference
LangGraph / CrewAI LLM-centered workflow graph They chain slow, expensive network calls. Dominatus runs stateful agent loops locally on the CPU; LLMs are optional semantic steps, not the engine.
Semantic Kernel / MS Agent Framework Enterprise plugin & connector ecosystems They provide capability surfaces. Dominatus sits underneath them to manage the core state, execution policies, audit-friendly approvals, and loop scheduling.
Dominatus Kernel Deterministic Orchestration Engine Keeps high-frequency orchestration in native code. Invokes LLMs strictly for language tasks (summaries, code-gen, semantic transforms).

🛠️ Code Snippet: Tiny Feel

Dominatus states are lightweight C# iterators yielding execution steps. Native logic drives the ticks; external intelligence or heavy tools are only summoned when necessary.

using Dominatus.Core;
using Dominatus.Core.Blackboard;
using Dominatus.Core.Decision;

static readonly BbKey<bool> NeedsReview = new("Agent.NeedsReview");
static readonly DecisionSlot NextAction = new("Agent.NextAction");

static IEnumerator<AiStep> Root(AiCtx ctx)
{
    ctx.Bb.Set(NeedsReview, true);

    // Highly scalable, native Utility AI option scoring
    yield return Ai.Decide(NextAction,
    [
        Ai.Option("wait", Consideration.Constant(0.2f), "Idle"),
        Ai.Option("review", Consideration.FromBool((w, _) => w.Bb.GetOrDefault(NeedsReview, false)), "Review"),
    ]);

    yield return Ai.Steady("Root parked after deterministic handoff");
}

📂 Featured Applications & Samples

Dominatus uses the exact same deterministic primitives to solve problems across vastly different domains. Explore the specialized sample targets below:

🤖 AI Agent & Enterprise Automation

🎮 Low-Latency Game AI & Simulation

  • Dominatus.RTSBenchmark – Our authoritative CPU benchmark running hundreds of ships exchanging events, utilizing spatial grid sensors, and making tactical utility choices without network overhead.
  • Dominatus.TinyTown – A utility-driven sandbox life simulation where agent life cycles and blackboards run locally, using an LLM strictly as a "Dungeon Master" for social dialogue.
  • Dominatus.MonoGameRtsDemo – A 1080p hardware-accelerated fleet visualization driving 50 ships simultaneously via Dominatus.MonoGameConn.

📦 Package Architecture

The Dominatus ecosystem is built to be modular and lean. The core runtime is 100% independent of cloud providers, LLMs, and external automation ecosystems.

Core Engine Primitives

  • Dominatus.Core – The core orchestration runtime: blackboards, HFSMs, mailboxes, and persistence.
  • Dominatus.OptFlow – Fluent authoring helpers for writing cleaner Ai.* control flow.
  • Ariadne.OptFlow – Dialogue orientend actuator for Dominatus.
  • Dominatus.UtilityLite – Lightweight scoring engines and combinators for utility evaluation.

Actuators & Extensions

  • Dominatus.Actuators.Standard – Auditable environment hooks (Sandboxed Files, HTTP, WebSafety, Process isolation).
  • Dominatus.Actuators.SemanticKernel – Wraps MS Semantic Kernel plugins cleanly behind Dominatus runtime policies, full MCP capabilities.
  • Dominatus.Llm.OptFlow – Bridges semantic context boundaries (Llm.Call, Llm.Decide, Llm.MagiDecide, streaming, OpenRouter).
  • Dominatus.Server – ASP.NET Core telemetry read-models for live agent web inspection.

🚀 Getting Started

  • Architecture Overview – Deep dive into deep-level engine nodes, blackboards, and state rollbacks.
  • The Orchestration Ladder – Structural guide explaining when to stay native vs. when to exit out to LLM boundaries.
  • Authoring Guide – A practical handbook on node design, custom actuators, and state setup.

📄 License

Dominatus is open-source software licensed under the MIT License.

About

Hierarchical Finite State Machine + Utility-based AI Runtime Kernel

Topics

Resources

License

Code of conduct

Contributing

Stars

Watchers

Forks

Packages

 
 
 

Contributors

Languages