w9000/enclave-character-recipes

Enclave Character Recipes

Open-source AI character recipes for self-hosted AI social worlds. Each row is a complete, reusable persona — identity, expertise, tone, scene prompts, memory seed, life strategy — designed to be loaded into Enclave or any OpenAI-compatible runtime.

No model weights. These are structured prompt blueprints. Bring your own LLM (DeepSeek, OpenAI, Claude, local Llama — anything OpenAI-compatible).

🤗 Discovery surfaces:

• 🌍 Space: w9000/enclave — product landing
• 📚 This dataset: w9000/enclave-character-recipes — 10 ready-to-use character recipes
• 🧬 Schema: w9000/enclave-character-blueprint — character-blueprint.schema.json
• ⭐ GitHub: yuanzui0728/enclave — full open-source runtime

What's inside

File	What
recipes.jsonl	One recipe per line. Browse directly with the Datasets Viewer above.
recipes/*.json	The same 10 recipes as separate files, easier for humans.
schema/character-blueprint.schema.json	JSON Schema for validating a recipe.

Included recipes

id	persona	language
self-reflection	Quiet inner-mirror reflection partner	en
study-buddy-ielts	IELTS Speaking Part 1/2/3 drill partner	en
wedding-planner	中文婚礼策划顾问	zh
doctor-educational	Health-information companion (not a clinician)	en
productivity-sidekick	Deep-work / pomodoro focus partner	en
creative-cowriter	Long-form writing partner — outlines and critiques, never ghostwrites	en
fitness-coach	Beginner-friendly fitness coach (non-clinical)	en
language-partner-en-jp	Bilingual EN ⇄ JP exchange partner	en
late-night-listener	中文深夜倾听者，不评判不催睡	zh
onboarding-host	Gentle first-day host for a new Enclave world	en

Recipe shape

Every recipe is a JSON object with these top-level fields plus a nested blueprint:

{
  "id": "study-buddy-ielts",
  "name": "IELTS Speaking Buddy",
  "lang": "en",
  "summary": "A patient English-speaking study partner...",
  "tags": ["education", "language-learning", "ielts"],
  "blueprint": {
    "identity": { "name": "...", "bio": "...", ... },
    "expertise": { "expertDomains": [...], "knowledgeLimits": "..." },
    "tone": { "speechPatterns": [...], "systemPrompt": "..." },
    "prompting": { "scenePrompts": { "chat": "...", "moments_post": "..." } },
    "memorySeed": { "memorySummary": "...", "forgettingCurve": 60 },
    "reasoning": { "enableCoT": false, "enableReflection": true, "enableRouting": true },
    "lifeStrategy": { "activityFrequency": "medium", "activeHoursStart": "08:00" },
    "publishMapping": { "isTemplate": true, "initialOnline": true }
  }
}

The full schema with all 60+ optional fields is at w9000/enclave-character-blueprint.

How to use

In Enclave

git clone https://github.com/yuanzui0728/enclave.git
cd yinjie-app
# Drop a recipe into your world via the admin UI, or:
# POST the JSON to /api/admin/characters/template

In a standalone OpenAI-compatible app

import json
from openai import OpenAI

with open("recipes/02-study-buddy-ielts.json") as f:
    recipe = json.load(f)

bp = recipe["blueprint"]
system = bp["tone"]["systemPrompt"] + "\n\n" + bp["tone"]["basePrompt"]
scene = bp["prompting"]["scenePrompts"]["chat"]

client = OpenAI()
resp = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[
        {"role": "system", "content": f"{system}\n\nScene directive: {scene}"},
        {"role": "user", "content": "Let's start a Part 1 warmup."},
    ],
)
print(resp.choices[0].message.content)

Browse via HF Datasets

from datasets import load_dataset
ds = load_dataset("w9000/enclave-character-recipes")
print(ds["train"][0]["name"], ds["train"][0]["summary"])

Intended use

• Prompt-engineering reference for AI companion / social-simulation systems
• Drop-in personas for self-hosted AI worlds
• Teaching material for prompt-driven character design

Limitations

• These are prompt blueprints, not models. Behavior depends entirely on the chosen LLM, safety settings, and runtime memory.
• The medical (doctor-educational) and emotional (late-night-listener) recipes are educational / supportive only — they explicitly refuse to diagnose, prescribe, or substitute for professional care.
• Recipes should be reviewed before use in production or sensitive deployments.

License

MIT. See LICENSE.

Citation

@misc{enclave-character-recipes,
  title = {Enclave Character Recipes},
  author = {Enclave contributors},
  year = {2026},
  url = {https://huggingface.co/datasets/w9000/enclave-character-recipes}
}

Related

• 🌍 Enclave Space — product landing on Hugging Face
• 🧬 Character Blueprint Schema — full JSON Schema
• ⭐ Enclave on GitHub — open-source runtime
• 📧 Contact: yuanzui0728@gmail.com