CodeExplore

CodeExplore is a core innovation that sets JeikCode apart from conventional AI coding assistants. JeikCode incorporates an ultra-low-overhead, change-aware weighted code graph engine featuring native Chinese and English natural-language semantic retrieval.

Whether tracking cross-file function calls and type definitions, or navigating fuzzy requirements described in mixed natural language and business terminology, CodeExplore accurately guides the Agent to the exact implementation sites.

---

Language & Ecosystem Support

JeikCode natively supports AST parsing and topological extraction for major programming languages:

Language / Framework	Parser	Supported Features
Rust	Tree-Sitter Rust	Structs, enums, trait impls, macro expansion contexts, module hierarchy
TypeScript / JavaScript	Tree-Sitter TS/JS	Classes, functions, arrow functions, interfaces, JSX/TSX exports and imports
Python	Tree-Sitter Python	Classes, functions, decorators, modules, dynamic references
Go	Tree-Sitter Go	Structs, interfaces, functions, package methods, receivers
Java	Tree-Sitter Java	Classes, methods, annotations, inheritance chains, interface impls
C / C++	Tree-Sitter C/CPP	Function prototypes, header references, classes, namespaces
C#	Tree-Sitter C#	Classes, namespaces, properties, async methods, pattern matching
Vue	Tree-Sitter Vue	<script setup> Composition API, template bindings, component dependencies
PHP / Ruby	Tree-Sitter PHP/Ruby	OOP structures, functions, class inheritance networks

---

Index Construction & Configuration Guide

1. Automatic Indexing & Real-Time Change Watching

Under normal circumstances within a project workspace, manual indexing is not required. The Agent automatically initializes the index on-demand; during daily development, the engine monitors file saving events and applies single-file incremental diff patches (1~3ms updates) without requiring manual rebuilds.

2. Multi-Repository Best Practices

If JeikCode is launched from a parent directory containing multiple independent repositories:

• It is strongly recommended to launch or build the index within each individual project directory;
• This prevents cross-project symbol collision and unnecessary indexing noise;
• Exception: Monorepos sharing a unified full-stack architecture (e.g. web/ and server/ within the same repository).

3. Manual CLI Commands

Run the following commands in the project root to analyze the repository and cache graph data:

# Initial or incremental index build
jeikcode init

# Force full re-scan and rebuild (clears previous cache)
jeikcode init --force

4. Integration with code_explore Tool

During task execution, the model queries the code graph using the built-in code_explore tool:

User: "Explain how user login JWT authentication and middleware interceptors work."
↓
Agent invokes: code_explore(path="src", query="user login JWT auth middleware")
↓
Code Graph:
1. Expands query via thesaurus to TokenValidator, verify_token, AuthMiddleware
2. Resolves call topology and extracts context code chunks
3. Aggregates upstream interceptor pipelines and downstream route handlers
↓
Agent returns exact execution timeline and file line references

You can ask high-level business or architectural questions without knowing exact filenames or symbol signatures in advance.

---

Code Graph Ignore System: .gitignore and .codegraphignore

To ensure that the code graph index maintains maximum signal-to-noise ratio and sub-millisecond retrieval speeds, the indexing engine operates with strict ignore and filtering mechanisms, supporting automatic .gitignore alignment, independent custom decoupling, and hardcoded defaults.

1. Default Binding to .gitignore

In standard Git repositories, the code graph crawler (driven by a high-throughput git ls-files pipeline) strictly adheres to the repository's .gitignore rules by default. External dependencies (node_modules/, target/), build outputs, temporary caches, and local configuration files are automatically skipped and never pollute the graph.

2. Deep Decoupling: Dedicated .codegraphignore

In enterprise codebases and team workflows, developers frequently encounter scenarios where certain files must be committed and tracked in Git, but should never be parsed or indexed by the code graph:

• Massive test datasets (Mock JSON, database fixtures, bulk SQL seeding scripts);
• In-tree vendor dependencies and minified UMD bundles (e.g. bundled UI libraries);
• Auto-generated protocol stubs (Protobuf outputs, massive ORM entity mappings);
• Bundled static binary assets or template files.

Adding these to .gitignore would prevent Git from tracking necessary code. JeikCode cleanly decouples code graph indexing from Git ignore rules via .codegraphignore:

• Core Mechanism: Any file or pattern declared in .codegraphignore will be skipped by the code graph indexer, even if it is actively tracked and committed in Git!
• Syntax Standards: Fully compatible with standard Glob and GitIgnore syntax (supports wildcards *, directory anchors /, and negative patterns !).

Configuration Locations

You can place .codegraphignore at any of the following locations (cascading automatically from local project to global scope):

1. Workspace Root (Recommended): <workspace>/.codegraphignore (applies to the current project)
2. Project Dot-folder: <workspace>/.jeikcode/.codegraphignore
3. Global User Configuration: ~/.jeikcode/.codegraphignore (applies globally across all local projects)

Example Template

Create or edit .codegraphignore and specify patterns to exclude from code graph parsing:

# ==============================================================================
# .codegraphignore — Code Graph & Symbol Index Ignore Rules
# ==============================================================================
# Regardless of whether files are tracked in Git, matching entries are excluded from CodeGraph.

# 1. Generated & Minified Assets
*.generated.*
*.g.cs
*.designer.cs
*.min.js
*.min.css
*.bundle.js
*.map
element-ui/
element-plus/

# 2. Frontend Dependencies & Build Caches
node_modules/
dist/
.output/
.next/
.nuxt/
.turbo/
.cache/
coverage/
*.tsbuildinfo

# 3. Python Virtual Environments & Bytecode
__pycache__/
*.py[cod]
.venv/
venv/
.pytest_cache/

# 4. Java / JVM Build Artifacts
.gradle/
*.class
*.jar
*.war

# 5. Rust Build Targets
target/
*.rlib

# 6. C / C++ / Native Binaries & Symbol Dumps
cmake-build-*/
*.o
*.obj
*.so
*.dll
*.exe
*.pdb

# 7. C# / .NET outputs (MSBuild config dirs only; a directory named bin can be source)
**/bin/[Dd]ebug/
**/bin/[Rr]elease/
obj/
TestResults/

# 8. Go / PHP In-tree Vendors
vendor/

# 9. IDEs & OS Metadata
.git/
.idea/
.vscode/
.DS_Store
Thumbs.db
*.log

Hot Reload: Changes to .codegraphignore take effect immediately on the next index rebuild (or whenever file saving triggers incremental watch, or via jeikcode init --force).

3. Built-in Hardcoded Safeguards

Even in projects lacking .gitignore or .codegraphignore, the JeikCode engine applies built-in defensive filters:

• Directory Skip List (SKIP_DIR_NAMES): Automatically prunes node_modules, target, obj, dist, build, .venv, vendor, coverage, and other build/dependency folders. A directory named bin is kept (Rust src/bin, Ruby and npm bin/ are source); bin/Debug and bin/Release are still pruned;
• Generated File Detection (is_generated_source): Skips *.designer.cs, *.g.cs, *.AssemblyAttributes.cs, *.min.js, *.bundle.js, *.map, and similar generated or minified files. Hand-written Properties/AssemblyInfo.cs stays in the index;
• Minified Web Bundle Interceptor (is_minified_web_bundle): Detects dense JS/CSS files (>32KB with <4 newlines in the first 4KB) to prevent Tree-Sitter AST blowups;
• Per-file Size Limits (max_index_file_bytes): 256KB ceiling for web scripts/styles and 768KB for general source code.