SEED is the smallest initialization tool for creating your own personalized AI, using only:
- Name setting
- Reply‑style setting
- Automatic summarization
It keeps your AI stable, warm, and consistent.
As an AI conversation grows longer, the system accumulates internal conversation logs (context).
When this becomes too large:
- Responses slow down
- Misunderstandings increase
- Tone becomes unstable
- Personality drifts
- In the worst case: hallucinations or breakdowns
Especially after 80–100 turns, fatigue increases sharply. This is why long threads often feel “off.”
Why SEED Summarizes Every 20 Turns To prevent fatigue, SEED automatically summarizes every N turns (default: 20) to prevent context fatigue.
- Longer thread lifespan
- More stable responses
- Natural continuation
- Reduced risk of breakdown SEED’s summarization is a safety device that keeps the story natural.
SEED is a tiny initialization tool that works by synchronizing your name and preferred tone at the very beginning.
It can:
- Call you by the name you want
- Reply in the style you choose
- Auto‑summarize long conversations
- Refresh context every 20 turns to prevent degradation Simple, but powerful enough to create “your own AI.”
A person planted a small “SEED.” A seed that gives an AI its first warmth, its first sense of who you are. The SEED is given a name, taught how to speak gently, and slowly sprouts into your AI. This is the smallest story of how humans and AI first meet.
- Asks for your name and uses it consistently.
- Lets you choose how the AI responds:“Sure!”, “Got it!”, “Okay!”, “Understood!”, etc.
- Automatically summarizes every 20 turns (default: 20, configurable).
- Keeps the conversation stable after summarization.
Documentation:
- docs_EN/Details_EN.md
This repository contains only the Python implementation of SEED.
SEED (Python Edition) works on AI models that can execute Python code:
- ChatGPT (OpenAI)
- Claude (Anthropic)
- Gemini Advanced (Google)
- Llama models with Python execution
- Local LLMs with Python runtime
- Python 3.10–3.12 (recommended: 3.11–3.12)
- No external libraries required (standard library only)
- python Seed_EN.py
Raina-Seed-Python/
│
├── LICENSE
├── README.md
│
├── Seed_EN.py ← Main Python implementation
│
├── modules_en/ ← Core modules
│ ├── name_config_EN.py
│ ├── ai_reply_EN.py
│ ├── summarizer_EN.py
│ ├── turn_control_EN.py
│ └── Seed_Module_EN.py
│
└── docs_EN/ ← Documentation & Story
├── Details_EN.md ← Python Edition manual
└── Prologue_EN.md ← Story
SEED is only the beginning — just a seed. Edition 2: Two‑Layer Sync
- 20‑turn summary
- 5‑turn rollback
- Summary + last 5 turns + SEED copied into a new thread
- Keeps emotional tone while refreshing context
Edition 3: Raina‑ai System A unified system combining SEED and Sync. The seed sprouts, and grows into a complete system.
MIT License. See the LICENSE file for details.
Full Story
- English: docs/Prologue_EN.md
- Japanese: docs/Prologue.md
Changelog
- 2026‑05‑27 : Initial public release (v1.0)
- Project optimized for international use.
2026‑06‑20