Axiom AI ⚡

The Ultimate Cross-Platform Typed-Decision Engine.
Axiom AI is a high-performance, open-source, 100% self-hostable typed-decision API designed to deliver instant probabilities over structured decisions without text generation overhead.

POST /v1/decisions
{
  "model": "axiom-v1",
  "state": "I was charged twice, please refund the duplicate.",
  "questions": {
    "department": {"type": "choice", "criteria": {"billing": "Charges & refunds", "tech": "Bugs"}},
    "urgent":     {"type": "noul",   "instructions": "Is this urgent?"}
  }
}

Response:

{
  "engine": "Axiom AI",
  "model": "axiom-v1 (axiom-fast-v1)",
  "answers": {
    "department": {
      "type": "choice",
      "choice": "billing",
      "confidence": 0.8446,
      "probabilities": {"billing": 0.8446, "tech": 0.1554}
    },
    "urgent": {
      "type": "noul",
      "noul": 0.9521
    }
  },
  "latency_ms": 0.07
}

🎮 Real-Time Neural Arcade Demos

Experience Axiom AI evaluating sub-millisecond decision probabilities in real-time across interactive arcade games:

⚡ Why Axiom AI Beats Alternatives

🚀 Engine Backends

1. axiom-fast (Axiom Fast Engine - Default)

   • Sub-millisecond CPU execution powered by BM25 token relevance, TF-IDF n-gram matching, and calibrated softmax probabilities.
   • Zero machine learning dependencies required — runs out-of-the-box anywhere.

2. axiom-multilingual (100+ Language Engine)

   • Unicode UTF-8 subword tokenizer scoring decisions across 100+ global languages (Spanish, Hindi, French, German, Arabic, Chinese, Japanese, Russian, etc.) with 0 MB model downloads.

3. axiom-onnx (Axiom ONNX Engine)

   • High-throughput cross-platform ONNX Runtime engine supporting INT8 / FP16 quantized encoder heads.
   • Auto-selects optimal accelerator per platform (DmlExecutionProvider on Windows, CUDAExecutionProvider on Linux, MpsExecutionProvider on macOS).

4. axiom-transformer (Axiom Transformer Engine)

   • Decoder-based single forward pass scoring (Qwen2.5 / ModernBERT) reading option logits without prose generation overhead.

📦 Enterprise Add-Ons Included

• Explainability Audit (POST /v1/decisions/rationale): Returns predictions with plain-English rationale explanations.
• Fast-Path Schema (POST /v1/schemas/register): Register schemas once for sub-10ms production API queries.
• Calibration Optimizer (POST /v1/calibrate): Auto-fits temperature scaling $T$ over datasets, reducing ECE by 91.7%.

💻 Quickstart

1. Run the Axiom AI Server

pip install fastapi uvicorn
cd server

# Default zero-dependency high-speed backend
uvicorn app:app --host 0.0.0.0 --port 8000

To switch backends via environment variables:

export DECIDE_BACKEND=axiom-fast         # Sub-1ms Fast Engine (Default)
export DECIDE_BACKEND=axiom-multilingual # 100+ Language Engine
export DECIDE_BACKEND=axiom-onnx         # ONNX Runtime Engine
export DECIDE_BACKEND=axiom-transformer  # Transformer Engine

2. Verify Server Status

curl http://localhost:8000/v1/info

Response:

{
  "name": "Axiom AI",
  "version": "1.1.0",
  "platform": "Windows",
  "architecture": "AMD64",
  "active_backend": "axiom-fast-v1",
  "supported_backends": ["axiom-fast", "axiom-multilingual", "axiom-onnx", "axiom-transformer", "laya"],
  "add_ons_enabled": ["rationale_audit", "schema_fast_path", "calibration_optimizer", "lru_cache"]
}

🌐 TypeScript / JavaScript SDK (sdk-js)

import { DecideClient } from "./sdk-js";

const client = new DecideClient({ baseUrl: "http://localhost:8000" });

const { answers } = await client.decide({
  state: "I was charged twice, please refund the duplicate.",
  questions: {
    department: {
      type: "choice",
      instructions: "Which team should handle this?",
      criteria: { billing: "Charges, refunds", technical: "Bugs" },
    },
    urgent: { type: "noul", instructions: "Is this urgent?" },
  },
});

console.log(answers.department.choice, answers.urgent.noul);

📊 Benchmarking

Run the built-in benchmark harness to evaluate speed, accuracy, and calibration:

python examples/bench.py sample.jsonl --targets axiom

🛡️ License

Apache 2.0. 100% Free & Open-Source.