XPDevs/Genesis-SPT-1.0

Genesis-SPT-1.0-NV25

A lightweight JSON conversational AI model. Works offline. No GPU required.

Overview

Genesis-SPT-1.0 maps user inputs to handcrafted responses. It does not use neural networks or inference engines. Just load the JSON and start chatting.

• —Model: Genesis-SPT-1.0-NV25 (Legacy modal)
• —Format: JSON (key to response)
• —Size: ~55 KB (249 entries)
• —Language: English
• —License: MIT

Features

• —Instant responses. No computation. Just a lookup.
• —Fuzzy matching. Handles typos, contractions, and paraphrases.
• —Reasoning simulation. Each response has an internal monologue using <|think|> tags. Clients can show this as chain-of-thought.
• —Fully offline. No internet or API calls needed.
• —Zero dependencies. Just a JSON parser and string matching.

Categories

The model covers 249 conversation patterns across 15+ categories.

Reasoning Tags

Every response includes a <|think|> tag with simulated reasoning.

{
  "hello": "<|think|>User is greeting me. Simple opener, probably just starting a conversation. I should respond warmly and invite them to share what they need.</|think|>Hey there! Genesis here, ready to help! What's on your mind?"
}

Clients can parse these tags. The thinking content can be shown before the clean response.

Usage (Python)

import json, re

with open("Genesis-SPT-1.0-original.json") as f:
    model = json.load(f)

def respond(input_text, model):
    input_lower = input_text.lower().strip()
    if input_lower in model:
        raw = model[input_lower]
        clean = re.sub(r"<\|think\|>.*?</\|think\|>", "", raw).strip()
        return clean
    best_key = None
    best_score = 0
    for key in model:
        if key in ("ver", "ModeInfo"):
            continue
        if key.lower() in input_lower:
            score = len(key) / len(input_lower)
            if score > best_score:
                best_score = score
                best_key = key
    if best_key:
        raw = model[best_key]
        clean = re.sub(r"<\|think\|>.*?</\|think\|>", "", raw).strip()
        return clean
    return "No match found."

print(respond("hello", model))

File Structure

Genesis-SPT-1.0-original.json   # Model data
README.md                       # This file

The JSON is a flat object.

{
  "ver": "Genesis-SPT-1.0-NV25 - Legacy modal",
  "hello": "<|think|>...</|think|>Hey there...",
  "bye": "<|think|>...</|think|>Take care...",
  ...
}

Limitations

• —No generative capabilities. Responses are handcrafted. The model cannot answer questions outside its dataset.
• —No memory. Each response is stateless. Conversation history is not considered.
• —English only. Designed for English conversation patterns.
• —Small scope. 249 entries cover common chat patterns but not deep knowledge.

License

MIT. Free to use, modify, and distribute.

Built by XPDevs.