nb-ai-context 1.6

非常强力的对任意it项目生成ai上下文单个markdown文档，丢给大模型或者知识库，非常强力，大模型幻觉大幅度减少。尤其是对python项目还有专门的ast语法树解析

nb_ai_context

English | 中文

🚀 An extremely powerful AI context generator - Merge any IT project into a single structured Markdown document for AI LLMs or RAG knowledge bases.

What is nb_ai_context?

nb_ai_context is not simply merging project file code, it is a context optimization tool specifically designed for AI code interaction, with the core goal of reducing AI illusions.

• nb_ai_context packages any IT project into a single markdown file for AI to learn and understand.

• You can search for "repomix" to understand its purpose - nb_ai_context generates documents that are far superior for AI learning, especially for Python projects.
  repomix simply merges multiple file contents, but nb_ai_context does much more than just merging.

• Why do you need nb_ai_context? Because third-party packages like google-genai, langchain, and pydantic have APIs that change too quickly. If you don't provide the latest documentation, AI will write outdated code using old package versions - sometimes even the imports won't work!
  You need to upload updated tutorial documents to AI so it can write correct code. Users shouldn't have to compromise by using old, outdated Python package versions just to use AI.

✨ Core Features

• ✅ AI Reading Guide - Adds instructions to help AI models understand document structure and reduce hallucinations
• ✅ File Dependencies Analysis - Analyzes import relationships, identifies entry points and core modules
• ✅ AST Metadata Extraction - Extracts class/function signatures from Python files without full source code
• ✅ Smart File Merging - Supports .gitignore, file filtering, directory exclusion
• ✅ Clear File Boundaries - Each file marked with project name and path for easy AI identification
• ✅ GitHub Project Support - Generate docs directly from GitHub zip URLs
• ✅ Chainable API - Elegant fluent interface for building context

📦 Installation

pip install nb_ai_context

Requirements:

• Python >= 3.7
• nb_path
• nb_log

🚀 Quick Start

Basic Usage (from examples/AiMdGenerator_example.py)

from nb_ai_context import AiMdGenerator

project_name = "nb_ai_context"
project_root = rf"D:\codes\{project_name}"

project_summary = f"""
- `{project_name}` is a powerful ai llm context generator library, it is used for ai llm and rag
- `AiMdGenerator(...)` is the main class to create ai context for llm.
"""

(
    AiMdGenerator(
        rf"D:\codes\nb_ai_context\ai_md_files_demo\{project_name}_all_docs_and_codes.md"
    )
    .set_project_propery(project_name=project_name, project_root=project_root)
    .ensure_parent()
    .clear_text()
    .add_ai_reading_guide()  # Add AI reading guide to help AI understand document structure
    .add_project_summary(
        project_summary=project_summary,
        most_core_source_code_file_list=[
            "nb_ai_context/__init__.py",
            "nb_ai_context/ai_md_generator.py",
            "nb_ai_context/contrib/gen_github_proj_ai_md.py",
        ],
    )
    .auto_merge_from_python_project_some_files()
    .show_textfile_info()
    .merge_from_dir(
        relative_dir_name='examples',
        use_gitignore=True,
        as_title=f"{project_name} examples",
        should_include_suffixes=[".py", ".md"],
        excluded_dir_name_list=[],
        include_ast_metadata=True,
    )
    .merge_from_dir(
        relative_dir_name=project_name,
        use_gitignore=True,
        as_title=f"{project_name} codes",
        should_include_suffixes=[".py", ".md"],
        excluded_dir_name_list=[],
        include_ast_metadata=True,
    )
    .show_textfile_info()
)

From GitHub Projects

from nb_ai_context import gen_github_proj_docs_and_codes_ai_md

gen_github_proj_docs_and_codes_ai_md(
    github_zip_url="https://codeload.github.com/fastapi/sqlmodel/zip/refs/heads/main",
    output_md_path=r"D:\ai_docs\sqlmodel_all_docs_and_codes.md",
    readme_file="README.md",
    docs_dir_name="docs",
    codes_dir_name="sqlmodel",
    should_include_suffixes=[".py", ".md"],
    excluded_dir_name_list=["tests", "__pycache__"],
)

📖 API Reference

AiMdGenerator Class

The core class for generating AI context. Inherits from NbPath and supports chainable calls.

Methods

Method	Description
set_project_propery(project_name, project_root)	Required first. Set project name and root directory
add_ai_reading_guide()	Add AI reading instructions to reduce hallucinations
add_project_summary(project_summary, most_core_source_code_file_list)	Add project summary with core file AST metadata
add_file_dependencies(file_list)	Analyze and add file dependency graph
auto_merge_from_python_project_some_files()	Auto-merge README.md, setup.py, pyproject.toml
merge_from_files(file_list, as_title)	Merge specific files
merge_from_dir(relative_dir_name, as_title, ...)	Merge entire directory with filters
merge_from_files_with_metadata(...)	Advanced merge with metadata control
show_textfile_info()	Display generated file statistics

merge_from_dir Parameters

.merge_from_dir(
    relative_dir_name="src",           # Directory relative to project_root
    as_title="Source Code",            # Section title in markdown
    project_root=None,                 # Override project root (optional)
    should_include_suffixes=[".py"],   # File extensions to include
    excluded_dir_name_list=[],         # Directories to exclude
    excluded_file_name_list=[],        # Files to exclude
    use_gitignore=True,                # Respect .gitignore rules
    dry_run=False,                     # Preview mode (no actual generation)
    include_ast_metadata=True,         # Include Python AST metadata
)

GitHub Helper Functions

Function	Description
gen_github_proj_docs_and_codes_ai_md(...)	Generate docs from GitHub repo with separate docs/codes directories
gen_github_proj_all_dirs_ai_md(...)	Generate docs from entire GitHub repo

🎨 Generated Markdown Structure

# 🤖 AI Reading Guide for Project: my_project
(Instructions for AI models)

# markdown content namespace: my_project project summary
(Project description)

## 📋 my_project most core source files metadata
(AST metadata for core files - no source code)

## 🔗 my_project File Dependencies Analysis
(Import relationships and dependency graph)

# markdown content namespace: my_project Source Code

## my_project File Tree (relative dir: `src`)
(Directory tree)

## my_project Included Files (total: X files)
(File list)

--- **start of file: src/main.py** (project: my_project) ---
### 📄 Python File Metadata: `src/main.py`
(AST metadata)

```python
(Full source code)
```

--- **end of file: src/main.py** (project: my_project) ---

🐍 Python AST Metadata Extraction

For Python files, automatically extracts:

• Module docstrings
• Import statements
• Class definitions (name, bases, decorators, docstring, methods, properties, class variables)
• Function definitions (name, parameters with types/defaults, return type, decorators, docstring)
• Constructor (__init__) details

🔒 Security

• Automatically respects .gitignore rules when use_gitignore=True
• Excludes hidden directories (starting with .)
• Supports manual exclusion of sensitive directories/files

🎯 Use Cases

1. AI Code Review - Let AI analyze entire project for quality, security, performance
2. RAG Knowledge Base - Import structured project docs into vector databases
3. Project Documentation - Generate comprehensive project overview for new team members
4. Learning Open Source - Quickly understand GitHub project architecture with AI assistance

🔗 Links

• GitHub: https://github.com/ydf0509/nb_ai_context
• PyPI: https://pypi.org/project/nb_ai_context/
• Issues: https://github.com/ydf0509/nb_ai_context/issues

📄 License

MIT License

---

nb_ai_context vs repomix: Professional Analysis on Reducing AI Hallucinations

• nb_ai_context is a byproduct of nb_path. AiMdGenerator inherits from NbPath, so it also supports infinite chainable operations, making it easy for users to chain-merge multiple folder sources into one markdown.
  However, nb_ai_context has now been separated out because generating AI context is harder, more complex, and requires more skill than file path operations.

• repomix is the top-tier third-party library for packaging IT project code into a single file, but nb_ai_context surpasses repomix in almost every aspect.

• nb_ai_context uses Python code with infinite chainable operations, supporting various methods - much more flexible than repomix's command-line approach. For example, it supports custom important AI prompt engineering.
  nb_ai_context allows users to specify the most important core file list via most_core_source_code_file_list, helping AI clearly understand the core APIs of third-party packages or your project. nb_ai_context Support adding custom AI prompt words through project_stummary input parameter.

• Users can verify whether nb_ai_context is really powerful or if the author is just bragging. The file ai_md_files_demo/nb_ai_context_all_docs_and_codes.md in this project was generated by nb_ai_context.
  You can upload nb_ai_context_all_docs_and_codes.md to Google AI Studio and let AI help you master how to use nb_ai_context - see if AI can learn how to use an obscure third-party package without prior training.

Core Design Philosophy Comparison

nb_ai_context

Designed specifically to reduce AI hallucinations, with multiple targeted features explicitly mentioned in the documentation:

• Detailed AI reading guide (explicitly tells AI how to understand document structure)
• Strict file boundary markers (clearly identifies start/end of each file)
• AST metadata extraction (lets AI understand code structure before seeing source code)
• Project dependency analysis (helps AI understand inter-module relationships)
• Forced path verification (requires AI to verify file paths exist when suggesting code changes)

repomix

Mainly focused on codebase aggregation, with the design goal of converting codebases into a single text file:

• Simple file separation markers
• Basic file filtering capability
• Preserves original code structure
• Lacks deep design specifically for AI understanding and reducing hallucinations

Key Feature Comparison for Reducing AI Hallucinations

Feature	nb_ai_context	repomix
AI Reading Guide	✅ Detailed guide explicitly telling AI how to understand document structure	❌ Basically none
File Boundary Identification	✅ Strict project name + path identification to prevent file confusion	⚠️ Simple file separators
Code Structure Preview	✅ AST metadata extraction (class/function signatures, docstrings)	❌ None, shows source code directly
Dependency Analysis	✅ Visualizes inter-module dependencies, helps AI understand architecture	❌ None
Core Entry Point Identification	✅ Clearly identifies core files and entry points	❌ None
Path Verification Requirements	✅ Explicit instructions requiring AI to verify file paths	❌ No explicit guidance
Project Summary	✅ Structured project overview helps AI quickly grasp key points	⚠️ Limited description capability
Hidden/Sensitive File Handling	✅ Supports .gitignore and manual exclusion of sensitive content	⚠️ Basic filtering

Practical Effect Comparison

When providing context generated by these tools to AI models:

nb_ai_context Advantages

1. Reduces file path hallucinations: By forcing AI to "check file paths" and "verify file paths exist in the File Tree", it nearly eliminates the problem of AI fabricating non-existent files
2. Reduces architectural misunderstanding: Through dependency graphs and AST metadata, AI more easily understands overall project architecture and won't incorrectly assume inter-module relationships
3. Precise code references: Strictly marked file boundaries enable AI to accurately reference specific files and line numbers when answering
4. More comprehensive context understanding: Project summaries and core file analysis help AI quickly grasp project focus instead of getting lost in details

repomix Limitations

1. Blurred boundaries: Simple file separators may cause AI to confuse content from different files
2. Lack of guidance: No explicit instructions on how AI should interpret document structure, increasing hallucination risk
3. Insufficient deep understanding: Directly exposes complete source code without providing code structure preview, making it difficult for AI to quickly grasp project architecture

Conclusion: nb_ai_context is Significantly Stronger at Reducing AI Hallucinations

nb_ai_context is not just a code aggregation tool, but a context optimization system specifically designed for AI-code interaction. It explicitly targets "reducing hallucinations" as a core goal, repeatedly emphasizing in the documentation:

⚠️ Important Notes

1. Do NOT hallucinate: Only reference code, classes, functions, and APIs that actually exist in this document
2. Check file paths: When suggesting code changes, always verify the file path exists in the File Tree
3. Respect the project structure: The File Tree shows the actual directory layout

While repomix is more of a general code aggregation tool without deep design specifically targeting AI hallucination issues. For scenarios requiring high-quality AI code understanding, review, or generation, nb_ai_context provides a more professional solution.

If you're preparing code context for AI systems, especially in enterprise applications or security-sensitive scenarios, nb_ai_context's professional design will significantly reduce the risk of AI producing dangerous hallucinations.

---

nb_ai_context - Let AI truly understand your code 🚀