kinrg 1.0.4

The Context Engine for AI Coding: Persistent incremental knowledge graph for token-efficient, context-aware reviews

Kinrg: The Intelligence OS for Code Context

"Stop letting AI guess. Let AI truly understand your codebase."

---

🌑 The Problem: AI is lost in your code

Most current AI tools read code linearly. In a large project (Monorepo), AI faces three hurdles:

1. Noise: Reading thousands of irrelevant lines, wasting tokens.
2. Oversight: Missing critical dependencies located in other files.
3. Hallucination: Providing incorrect solutions due to a lack of overall architectural understanding.

🔥 The Breakthrough: Graph Intelligence Layer

kinrg is a Context Engine that transforms your source code into a living Knowledge Graph. It allows your AI Agent to "see":

• Blast Radius: Precisely identify the impact zone of any code change.
• Logic Flows: Understand how data flows from entry points to core logic.
• Smart Pruning: Automatically strip away 95% of context noise so AI focuses on the right issue.

---

💎 Core Pillars

🚀 ULTIMATE PERFORMANCE	🧠 DEEP AST TECHNOLOGY	🌐 OPEN ECOSYSTEM
Save up to 27x Tokens through intelligent context filtering, sending only what truly matters.	Powered by Tree-sitter to understand source code like a real compiler (19+ languages).	Automatic MCP configuration for Cursor, Claude Code, Windsurf, Zed, Continue, and OpenCode.
Updates in < 2s even for projects with thousands of files, thanks to incremental algorithms.	Hybrid Search: The perfect blend of Keyword (FTS5) and Semantic Vector Search (Embeddings).	Deeply integrated into Anthropic's Model Context Protocol (MCP).

---

🔌 Supported AI Ecosystem

With a single kinrg install command, you can upgrade the intelligence of: Cursor • Claude Code • Windsurf • Zed • Continue • OpenCode • Antigravity

---

⚙️ Installation & Setup

1. Requirements

• Python: Version 3.10 or higher.
• IDEs: Best compatible with tools supporting MCP (Cursor, Claude Code, etc.)

2. Installation Methods

You can install kinrg flexibly via popular package managers:

# Standard installation via pip
pip install kinrg

# Install via pipx (Recommended to avoid dependency conflicts)
pipx install kinrg

# Use uv for maximum installation performance
uv tool install kinrg

3. Advanced Features (Optional)

Unlock the full power of kinrg by installing extension modules:

# Enable Semantic Search (Vector Embeddings)
pip install "kinrg[embeddings]"

# Enable Community & Module Structure Analysis
pip install "kinrg[communities]"

# Enable Automated Wiki Documentation (requires LLM summaries)
pip install "kinrg[wiki]"

# Install all features in a single command
pip install "kinrg[all]"

4. Connect to your AI Editor

Use the install command to let kinrg automatically detect your installed IDEs and configure the MCP Server:

kinrg install

Note: To configure for a specific platform, use: kinrg install --platform [cursor|claude-code|zed|...].

---

💡 First-time Usage Guide

To get started, open your project in an IDE (like Cursor or VS Code) and follow these steps:

1. Build the Graph: Type in your AI chat: "Build the source code graph using kinrg".
2. Review Code: When you have new changes, ask: "Review my latest changes using kinrg".
3. Search: "Find all functions related to login using semantic search".

kinrg will run in the background, providing all necessary data for the AI to work with maximum accuracy.

---

🛠️ Comprehensive Intelligence: 26 MCP Tools

Your AI Agent will automatically utilize these 26 specialized tools to assist with every programming task:

🔍 Analysis & Review (Review & Impact)

• audit_diff_tool: Risk analysis and precise change boundary identification based on Git diff.
• trace_impact_tool: Precisely identify the "impact radius" of code changes (Blast Radius).
• prepare_review_tool: Provide optimized code snippets for AI review, saving tokens.
• quick_brief_tool: Retrieve hyper-compressed context (~100 tokens) for any task.
• check_complexity_tool: Detect functions/classes that are too complex based on line count.

🌐 Search & Graph (Search & Query)

• smart_search_tool: Search code using natural language or meaning (Vector Search).
• inspect_relationships_tool: Query Callers, Callees, Imports, Tests, and Inheritance.
• global_search_tool: Search source code across multiple registered projects simultaneously.
• fetch_stats_tool: View detailed statistics on node counts, edges, and languages.

🌊 Execution Flow

• explore_flows_tool: List major execution flows (Entry points) sorted by criticality.
• detail_flow_tool: Step-by-step trace of a specific source code flow.
• check_damaged_flows_tool: Find user flows directly affected by the code you just modified.

🏛️ Architecture & Community

• generate_blueprint_tool: Overview architecture map and coupling dependency warnings.
• cluster_modules_tool: Automatically cluster source code into logic modules (Leiden algorithm).
• inspect_module_tool: View the boundaries and members of a specific code cluster.

📝 Documentation & Resources (Docs & Wiki)

• build_wiki_tool: Automatically generate a full Markdown Wiki for the project from the graph.
• read_wiki_tool: Retrieve documentation content for a specific module.
• read_manual_tool: Quickly access technical documentation sections for kinrg.

🔧 Refactor & Database

• refactor_advisor_tool: Preview wide-reaching renames and detect dead code for cleanup.
• execute_refactor_tool: Execute refactoring changes that have been previewed.
• catalog_database_tool: List all database tables (SQL, Prisma, Django, etc.).
• inspect_table_tool: Table schema details, foreign keys, and code files querying it.

🏗️ Administration (Core & Registry)

• sync_knowledge_tool: Initialize or update the project architectural graph (Full/Incremental).
• optimize_graph_tool: Re-run post-processing steps (Flows, Communities, FTS) for the graph.
• vectorize_codebase_tool: Compute vector embeddings for the entire codebase to enable semantic search.
• manage_registry_tool: Manage and list repositories registered in the multi-repo system.

---

💻 CLI Reference

Control kinrg directly from your terminal using the following command suite:

Command	Function
kinrg install	Auto-detect and configure MCP for IDEs.
kinrg index	Scan and build the architectural graph for the first time (Full parse).
kinrg sync	Fast update of the graph (Processes only changed files).
kinrg report	View detailed statistics on nodes, edges, and graph health.
kinrg monitor	Watch mode: Auto-update the graph immediately upon file save.
kinrg map	Launch the web interface to visualize the project graph.
kinrg docs	Automatically generate a Markdown Wiki based on code structure.
kinrg audit	Risk analysis of current changes compared to Git.
kinrg attach <path>	Register a new project into the multi-repo management system.
kinrg registry	List all repositories registered in the Registry.
kinrg serve	Start the MCP server to connect with AI Agents.

---

⚡ Slash Commands

When chatting with MCP-enabled AIs (like Cursor, Claude Code, OpenCode), you can use these shortcuts:

• /kinrg:build-graph: Ask AI to update the entire project architectural map.
• /kinrg:review-delta: AI will analyze and comment on your latest code changes.
• /kinrg:review-pr: Perform a full PR review process with deep risk analysis.

---

📝 License

MIT. All source code data is kept secure locally within the .kinrg/ directory.

Website • GitHub

_{Released under the MIT License. All source code data is kept secure locally within the .kinrg/ directory.}