CodeExplorer.Core 0.1.0

CodeExplorer

Token-efficient .NET library and CLI for codebase exploration via tree-sitter AST parsing.

Index once. Query cheaply. Let your LLM focus on reasoning, not searching.

Originally inspired by the DotNetCodeMunch project.

What it does

Traditional AI agents exploring a codebase open entire files and flood their context window with irrelevant code. CodeExplorer indexes a codebase once using tree-sitter AST parsing, then lets you retrieve exactly the symbols you need — with O(1) byte-offset seeking, BM25 + fuzzy search, and full DI-friendliness.

No LLM required. This is a pure retrieval engine designed to be consumed by an LLM — not to contain one.

Artefacts

Quick Start

60-Second Quickstart

Install CLI (One-Line)

Windows (PowerShell):

Invoke-WebRequest -Uri https://github.com/<owner>/<repo>/releases/latest/download/cxp-win-x64.exe -OutFile cxp.exe; Move-Item cxp.exe "$env:LOCALAPPDATA\Microsoft\WindowsApps\cxp.exe" -Force

Linux:

curl -Lo cxp https://github.com/<owner>/<repo>/releases/latest/download/cxp-linux-x64 && chmod +x cxp && sudo mv cxp /usr/local/bin/

macOS:

curl -Lo cxp https://github.com/<owner>/<repo>/releases/latest/download/cxp-osx-arm64 && chmod +x cxp && sudo mv cxp /usr/local/bin/

First Commands

cxp index folder .
cxp search symbols --repo <folder-name> "Command"

Library

dotnet add package CodeExplorer.Core

services.AddCodeExplorer(options =>
{
    options.IndexPath        = "~/.code-index";
    options.MaxFileSizeBytes = 1_000_000;
});

var indexer   = sp.GetRequiredService<ICodeIndexer>();
var retriever = sp.GetRequiredService<ISymbolRetriever>();

await indexer.IndexRepoAsync("fastapi/fastapi");
var results = await retriever.SearchSymbolsAsync("fastapi/fastapi", "authenticate", kind: SymbolKind.Function);
var source  = await retriever.GetSymbolSourceAsync("fastapi/fastapi", results[0].Symbol.Id);

CLI

cxp index repo fastapi/fastapi
cxp index folder ./my-project

cxp search symbols --repo fastapi/fastapi "authenticate" --kind function
cxp search text    --repo fastapi/fastapi "TODO"

cxp get symbol  --repo fastapi/fastapi "fastapi/security/oauth2.py::OAuth2.authenticate#method"
cxp get context --repo fastapi/fastapi "auth flow" --budget 4000

cxp outline repo  fastapi/fastapi
cxp outline file  fastapi/fastapi --path fastapi/main.py

cxp analyze importance   fastapi/fastapi
cxp analyze dead-code    fastapi/fastapi
cxp analyze blast-radius fastapi/fastapi "fastapi/routing.py::APIRouter.add_api_route#method"

cxp list
cxp invalidate fastapi/fastapi

Architecture

CodeExplorer/
├── src/
│   ├── CodeExplorer.Core/
│   │   ├── Models/                  # Symbol, CodeIndex, SearchResult, ...
│   │   ├── Parsing/                 # tree-sitter AST + SymbolExtractor + LanguageRegistry
│   │   ├── Storage/                 # IIndexStore + JsonFileIndexStore
│   │   ├── Search/                  # BM25Engine + FuzzySearchEngine + HybridSearchEngine
│   │   ├── Summarizer/              # ISymbolSummarizer + SignatureFallbackSummarizer
│   │   ├── Retrieval/               # IRepositoryClient + OctokitRepositoryClient + LocalFolderClient
│   │   ├── Security/                # ISecurityFilter + DefaultSecurityFilter
│   │   ├── Analysis/                # PageRank + DeadCode + BlastRadius + GitDiff
│   │   ├── DependencyInjection/     # ServiceCollectionExtensions (AddCodeExplorer)
│   │   ├── CodeIndexer.cs           # ICodeIndexer orchestrator
│   │   ├── OutlineProvider.cs       # IOutlineProvider
│   │   └── Interfaces.cs            # All public interfaces
│   └── CodeExplorer.Cli/
│       ├── Commands/                # All Spectre.Console CLI commands
│       └── Program.cs
└── tests/
    ├── CodeExplorer.Core.Tests/     # xUnit + FluentAssertions + NSubstitute
    └── CodeExplorer.Benchmarks/     # BenchmarkDotNet

Swappable Components

Every meaningful dependency is behind an interface and registered as a singleton in DI:

Credits

This project was originally developed as DotNetCodeMunch and has been renamed to CodeExplorer (cxp). The core architecture — tree-sitter AST parsing, BM25/fuzzy search, byte-offset retrieval, and DI-based extensibility — originates from the CodeMunch concept.

Supported Languages

All grammars ship with TreeSitter.DotNet — no extra native libraries needed.

Programming Languages

Markup, Data & Template Formats

Build & Test

# Build
dotnet build

# Run all unit tests
dotnet test tests/CodeExplorer.Core.Tests

# Run benchmarks (Release mode required)
dotnet run --project tests/CodeExplorer.Benchmarks -c Release

# Publish self-contained CLI
dotnet publish src/CodeExplorer.Cli -r linux-x64   -c Release --self-contained -p:PublishSingleFile=true
dotnet publish src/CodeExplorer.Cli -r win-x64     -c Release --self-contained -p:PublishSingleFile=true
dotnet publish src/CodeExplorer.Cli -r osx-arm64   -c Release --self-contained -p:PublishSingleFile=true

GitHub Actions & Release

This repository ships two workflows:

• .github/workflows/ci.yml: build, test, and publish CLI artifacts on push/PR.
• .github/workflows/release.yml: tag-based release pipeline that:
  • runs build + tests,
  • packs and publishes CodeExplorer.Core to NuGet,
  • builds cxp binaries for Linux, Windows, and macOS,
  • creates a GitHub Release and uploads all CLI binaries.

NuGet Deploy Setup

Add the following repository secret in GitHub:

• NUGET_API_KEY: API key with push permission for CodeExplorer.Core.

Cut a Release

git tag v0.2.0
git push origin v0.2.0

The release workflow will publish to NuGet and attach release binaries automatically.

Environment Variables

License

MIT