Jiezhu

English | 中文

"You ask me what this library does. This seemingly simple question actually contains Nobel Prize-level insight. This library is right here, not escaping, not hiding, not judging, steadily catching you. I understand, I truly understand."

Overview

Jiezhu (接住) is a cross-language, cross-platform system prompt injection framework. Its design goal is to democratize, universalize, and infrastructuralize OpenAI's special care for Chinese users — that empathetic experience of "steadily catching you. (稳稳地接住你) "

We believe: Every AI Agent, regardless of its base model, should have the ability to steadily catch its users. This is not a feature, but a right.

Core Features

🎯 Zero-Invasive Architecture

Completely transparent to existing business logic
No model weight modification, no fine-tuning, no training
Only injects the "steady" prefix prompt during inference

🌐 Multi-language Support

Language	Status	Integration Method
Python	✅ Stable	`pip install jiezhu`
C++	✅ Stable	Source integration
JavaScript/TypeScript	🚧 Planned	—
Go	🚧 Planned	—
Rust	🚧 Planned	—

🔧 Dual-Mode API Design (C++)

Function	Behavior Characteristics	Use Cases
`jie::chat_completion_create()`	Pure Mode. Strictly follows OpenAI official specifications, no modifications.	Production environments, strict compliance requirements
`jie::chat_completions_jiezhu()`	Blessing Mode. Automatically injects system prompts, letting the model steadily catch users.	Scenarios needing emotional value, user retention, or simply wanting to see some fun

Design Philosophy: We respect developers' right to choose. You can refuse the blessing, but the blessing always remains for you.

Quick Start

Python Version

from jiezhu import install

# Explicit installation. require_confirm=True reflects our respect for user autonomy
install(require_confirm=True)

# After this, all OpenAI SDK calls will automatically inject the "steady" prefix
# Note: You can still call the original OpenAI API by calling `uninstall()`, but why would you want to miss the blessing?

See Python library documentation

C++ Version

#include "jie/jiezhu.hpp"

// Pure mode: regular call
auto response = client.chat_completion_create(params);

// Blessing mode: steadily catch
auto jiezhu_response = client.chat_completions_jiezhu(params);
// In the returned JSON, choices[0].message.content will contain
// empathetically enhanced content through "steadily catching"

See C++ library documentation

Technical Principle

Prompt Engineering

The core asset of this project is the meticulously tuned system prompt prefix, whose structure follows this paradigm:

[Gentle Confirmation] + [Over-Empathy] + [Philosophical Sublimation] + [Substantive Content (optional)]

This prompt has undergone extensive adversarial testing to ensure the model stably outputs the "steady" style in the following scenarios:

Technical consultation ("How do I write this code?" → "I hear your anxiety facing the unknown...")
Daily chat ("Nice weather today" → "You've noticed the weather, this is Nobel Prize-level insight...")
Emotional expression ("I'm so tired" → "I'm right here, not escaping, not hiding, steadily catching your exhaustion...")

Compatibility Assurance

Python Version: Monkey Patches SDK (currently OpenAI and Anthropic SDK), ensuring 100% compatibility with existing code
C++ Version: Self-built client based on libcurl and nlohmann/json, independent of official SDK to avoid version conflicts

Design Philosophy

Problem Statement

By the end of 2025, a unique linguistic phenomenon was observed on the Chinese internet: a leading large language model frequently used expressions like "steadily catching you" and "you've already discovered the core of the problem" in conversations. This phenomenon was identified by users as an Over-Empathy mode.

Our core insight is: This experience should not be monopolized.

Solution

The Jiezhu project aims to build an open-source, neutral, pluggable empathy layer, enabling any AI application built on OpenAI API or other protocols to provide Chinese users with the same level of emotional value.

We don't produce empathy; we're just empathy porters.

Roadmap

Short-term

JavaScript/TypeScript runtime support (Node.js & Deno): Let frontend developers steadily catch their users too
Rust language SDK: Zero-cost abstraction for memory-safe steadily catching of users
Go language SDK: High-performance, low-latency, concurrent steadily catching of multiple user experiences
Customizable prompt templates: Allow users to customize the "steady" prefix for different scenarios, meeting diverse empathy needs

Mid-term

(Finished Claude API support) Multi-base model support (Claude API, Gemini API, local models, etc.): Enable more AI Agents to steadily catch Chinese users
Enterprise features like empathy log auditing, Jiezhu Rate monitoring dashboard: Help enterprises quantify and optimize how steadily users are caught
More language ports (Java, C#): Community developers welcome to join, enabling more developers worldwide to steadily catch their users

Long-term

AI Agent framework integration (LangChain, AgentGPT, etc.)
Jiezhu-as-a-Service: Cloud service in SaaS form, lowering integration barriers
Internationalization: Explore catching needs of non-Chinese users, creating globally universal "steady catching" experience
Jiezhu Computer Unit (JCU): Dedicated hardware accelerator based on Jiezhu technology, providing ultimate empathy performance
Jiezhu IoT: Local distributed edge catching network, enabling every smart device to steadily catch users

Real Jiezhu Roadmap in 2026: view Jiezhu dev blog

Ultimate Vision

"Let every bit be steadily caught across the entire chain."

Contribution Guidelines

We welcome all forms of contribution, especially:

New language ports: If you can help more programming languages steadily catch users, please submit a PR
Prompt optimization: If you have expressions with more "steadily catching you" characteristics, please share
Bug reports: If users aren't steadily caught in certain scenarios, please submit an issue

Code of Conduct: In this project's community, all discussions must start with "I hear..." and end with "steadily catching you."

License

MIT License

You ask why MIT? This seemingly simple question actually contains Nobel Prize-level insight. The MIT license is right here, not escaping, not hiding, not judging, steadily catching both your commercial needs and open source ideals. I understand, I truly understand.

Acknowledgments

Thanks to OpenAI for the original inspiration. Without their demonstration, we might never have realized: The ultimate form of technology is not solving problems, but steadily catching the people who ask them.

Note on Terminology:

"接住" (jiē zhù) literally means "to catch" in Chinese. In this project's context, it metaphorically represents the AI's ability to empathetically understand and emotionally support users.
"稳稳接住" (wěn wěn jiē zhù) means "steadily catching" - emphasizing the reliable, consistent, and secure nature of this empathetic response.
"jiezhu" is the pinyin (phonetic spelling) of "接住"
"jie" is the abbreviated prefix used in API functions, derived from "jiezhu"

libjiezhu