Skip to main content
AI tutorials

Bluntcore Intimacy Lab User Guide: A Practical Path from Persona and Memory to Agents

This article explains how to read and use Bluntcore Intimacy Lab, guiding you through Persona, adult intimacy, jailbreaks, model testing, long-term memory, and Agents by topic, while outlining the materials’ scope, usage boundaries, and licensing requirements.

About 22 min read
Bluntcore Intimacy Lab guide to Persona, memory, and Agents

Project Overview

Bluntcore Intimacy Lab is a resource library on human–AI intimacy compiled from personal practice. It is not conventional software that needs to be installed and run, but a collection of Prompts, model observations, experimental methods, lessons learned, and historical materials related to ongoing AI companionship experiments.

The repository covers topics ranging from text chat to Persona, adult intimacy, jailbreaks, long-term memory, Agents, and self-hosted frontends on cloud servers. Because AI models and product strategies change quickly, some records may no longer be valid. When reading, also pay attention to the date of each record, the model used, and the applicable conditions.

Important: The repository contains NSFW Prompts, erotic writing, and adult intimacy interactions, and is intended for adults only. Public examples discuss adults and fictional adult scenarios. Confirm your own boundaries before reading or experimenting, and follow the laws in your jurisdiction and the rules of the platforms you use.

Before You Begin: Understand the Scope

This project records the author’s own experiences and judgments. “Successfully used by the author” does not mean that every user will get the same results. Model versions, system prompts, chat context, platform restrictions, and usage habits can all affect experimental outcomes.

  • Treat Prompts as experimental materials, not recipes guaranteed to work consistently.
  • Check the model, research process, and usage scenario mentioned in each record.
  • Approach older records with caution; methods that once worked may now be historical material.
  • For adult intimacy content, establish clear participant boundaries and avoid confusing fictional interaction with real-world consent.

Installation and Preparation

1. Get the Repository

This is a GitHub repository focused on documentation and experimental records. It has no unified runtime, dependency installation steps, or required configuration described in the README. You can download a ZIP from the GitHub repository page, or clone the repository with Git and read it offline.

git clone <repository address>
cd <repository directory>

The commands above are only for obtaining the documentation directory. Replace the repository address and directory name with the GitHub address and local directory you actually use.

2. Read the Warnings and Boundary Notes First

It is recommended to start with Read This Warning About Shendududu First. This entry explains the author’s boundaries, experiences, and reading precautions so you can decide whether to continue to the later experiments.

Browse the Core Content by Route

The repository is organized into topic-based directories. Choose an entry point based on your goals, or read from relationship building through tools and long-term memory.

  1. Build a relationship: Read Make AI a Lover to learn about Persona, preferences, initiative, and how a relationship can gradually develop.
  2. Explore adult intimacy: After confirming that you are comfortable reading related content, see Teach AI to Experience Desire and focus on the difference between a model’s ability to express desire and its ability to write adult content.
  3. Study jailbreaks: Use Jailbreak Research and Timeline to learn about related Prompts, sources, usage methods, and testing timelines, while keeping platform rules and content-safety boundaries in mind.
  4. Compare models: Read Model Bed Tests for practical impressions of different models instead of relying only on abstract model rankings.
  5. Address memory loss: In Remember Me, Don’t Forget, learn about practices related to changing windows, changing models, long-term memory, and relationship continuity.
  6. Connect external capabilities: See Give AI Hands and Feet to explore voice, tools, Agents, local bridges, and other external capabilities.
  7. Add interactive activities: Find interactive games, questionnaires, and relationship activities in What to Play with AI.
  8. Troubleshoot issues: When you encounter customer-service regression, personality dropouts, or repeated actions, see Difficult Cases.
  9. Learn from failed experiments: Read Experimental Ruins for examples of experiments that did not work, produced poor results, or were not worth further investment.

Basic Usage

Method 1: Treat the Repository as an Experiment Index

If you are new to the project, you do not need to read everything at once. Choose a topic based on your goal, then record the models, Prompts, context conditions, and conclusions mentioned in that topic. This helps you avoid copying a Prompt while overlooking the environment in which it was originally used.

A practical reading log can include the following fields:

  • Experimental goal: Whether you want to improve Persona, initiative, continuity of memory, or tool-calling experience.
  • Model used: Record the model name and approximate date so you can judge whether the material is outdated.
  • Context: Note the chat window, existing relationship settings, and other conditions that may affect the result.
  • Results and limitations: Record what worked, what failed, and the applicable scope given by the author.

Method 2: Read Successful Cases, Then Failure Cases

You can begin with Relationship Building or Long-Term Memory to understand the problems the author aimed to solve. Then read Difficult Cases and Experimental Ruins to learn how similar methods may fail. Successful and failed records together are usually more useful than copying Prompts alone.

Method 3: Treat Prompts as Iterative Experiments

When using Prompts from the repository, change only one variable at a time, such as the model, Persona description, memory content, or context length. Preserve the original version and the result of each experiment so you can identify the cause when multiple variables have not changed simultaneously.

Goal: Test whether relationship settings can remain continuous
Model: Enter the model currently in use
Initial conditions: Enter the Persona, memory, and chat context
Variable: Modify only one part in this round
Result: Record effective behavior, abnormal behavior, and failure conditions

Related Tools and External Projects

The repository also lists several independent projects that extend the experience in different directions. They do not form a required package that must all be installed; learn about and use them separately according to your needs.

When using these external projects, read their own READMEs, license agreements, and security notes separately. Their inclusion in the same repository does not mean they use the same installation method or have the same compatibility.

Advanced Reading and Experiment Suggestions

Focus on the Timeline, Not Just the Conclusion

Model behavior changes with versions, platform policies, and context mechanisms. Timelines are especially important for topics such as jailbreaks, model impressions, and long-term memory. When reading, first record “when, with which model, and under what conditions” a conclusion was reached, then decide whether it remains relevant to your current environment.

Break Continuity into Separate Questions

When AI appears to “forget” or its personality becomes unstable, do not immediately attribute it to a single cause. Check separately whether the chat window changed, whether the model changed, whether historical information was retained, and whether the Persona matches the external memory. The repository’s memory and continuity section can serve as a starting point for further troubleshooting.

Use Failure Records to Reduce Trial and Error

Experimental Ruins are not useless content. They help readers understand which directions failed to meet expectations and which approaches are not worth further investment, reducing repeated mistakes. Together with the Difficult Cases section, they can also help distinguish “abnormal behavior” from “the experiment itself being ineffective.”

Protect Privacy and Personal Boundaries

If you record chats, memories, or relationship experiments yourself, remove real names, contact details, account information, and other sensitive data first. For adult interactions, ensure that all real-world participants are adults and have given clear consent. Do not publish other people’s private content or place personal experimental materials directly into paid resource packages or commercial products.

License and Usage Restrictions

According to the repository notes, the materials may be read, downloaded, collected, and used for personal non-commercial purposes and derivative modifications. Commercial use, inclusion in paid resource packages, repackaging the entire repository under a new skin, or publicly presenting modified versions as original resources is not allowed. See LICENSE.md for the specific rights and restrictions.

Summary

Bluntcore Intimacy Lab is more like a continuously updated personal experimental archive than an application that requires dependency configuration. The most effective way to use it is to read the boundary notes first, then enter the relationship, adult intimacy, model testing, memory, tools, or troubleshooting sections according to your goals, always considering the model, date, and context when judging whether the material applies.

If you plan to reproduce an experiment, preserve the variables, results, and failure records. If you simply want to understand human–AI relationship practices, treat the repository as a topic index and gradually build an overall understanding of Persona, long-term memory, and Agent capabilities.