AI Code Context Collector

Local-first code context extractor for web AI assistants (ChatGPT, Claude, Gemini, GitHub Copilot Web).

🚀 Overview

AI Code Context Collector is a production-quality VS Code extension that automatically analyzes large frontend and backend codebases to extract precise, character-accurate code context for a given task.

Instead of manually copy-pasting dozens of files or uploading entire massive repositories to web-based AI tools, you describe your task in plain English. The extension analyzes dependencies, identifies relevant files and symbols, extracts the exact local source code, and generates a clean, self-contained ai-task-context.md file ready to be uploaded or pasted into ChatGPT, Claude, Gemini, or GitHub Copilot Web.

🔒 Privacy & Local-First Architecture

• 100% Local: Source code is analyzed locally on your machine.
• No Cloud Uploads: The extension NEVER uploads your source code to OpenAI, Anthropic, Google, GitHub, or any external servers.
• No Telemetry: Zero tracking, analytics, or telemetry.
• No Project Modifications: The extension only creates ai-task-context.md in your workspace root. It never modifies your existing .ts, .js, .html, or .json files.
• Automated Secret Redaction: Detects API keys, private keys, database connection strings, and tokens, automatically redacting them before packaging.

🏗️ Two-Stage Architecture

                    User Task
                       │
                       ▼
             ┌────────────────────┐
             │ Local Static Scan  │
             │                    │
             │ Files & Imports    │
             │ AST Symbols        │
             │ Angular Decorators │
             │ Express Routes     │
             │ HTTP APIs & Models │
             └─────────┬──────────┘
                       │
                Candidate files
                       │
                       ▼
             ┌────────────────────┐
             │ Optional Ollama    │
             │ Local Small LLM    │
             │                    │
             │ Relevance ranking  │
             │ Code block select  │
             └─────────┬──────────┘
                       │
                Selected symbols
                       │
                       ▼
             ┌────────────────────┐
             │ Exact Source       │
             │ Extraction         │
             │                    │
             │ ORIGINAL CODE      │
             │ NO REWRITING       │
             └─────────┬──────────┘
                       │
                       ▼
             ┌────────────────────┐
             │ Markdown Generator │
             └─────────┬──────────┘
                       │
                       ▼
              ai-task-context.md

Critical Guarantee: No AI Hallucinated Source Code

The local LLM is used strictly for ranking candidate files and symbols. The extension itself directly reads your local files from disk and copies the exact, character-accurate source code with surrounding class context, imports, and dependencies.

🦙 Ollama Integration (Strictly Optional)

Ollama is an optional local enhancement. The extension works out of the box with deterministic static analysis even if Ollama is not installed or running.

Recommended Local Model

ollama pull qwen2.5-coder:1.5b

Modes of Operation

1. Mode A — Static Mode (Ollama Offline / Disabled):
   • Uses AST scanning, keyword matching, relationship graph distance, open file boost, and active selection boost.
2. Mode B — AI-Assisted Mode (Ollama Running):
   • Sends candidate metadata (file paths, symbols, routes, API calls) to qwen2.5-coder:1.5b running on http://localhost:11434.
   • Parses structured JSON response and ranks top relevant symbols.
   • If Ollama fails, times out, or returns invalid schema, it gracefully falls back to Static Mode without interrupting your workflow.

💡 Supported Technologies

• Frontend: Angular (@Component, @Injectable, @Directive, @NgModule, @Pipe, .html templates, .scss/.css styles), TypeScript, JavaScript, HTML, SCSS, CSS.
• HTTP APIs: Angular HttpClient, axios, fetch with endpoint resolution.
• Backend: Node.js, Express.js (router.get, app.post, route-to-controller mapping, middleware).
• Database & ORM: MongoDB, Mongoose (mongoose.model, Schema), MySQL/SQL models.

🎮 Commands

⚙️ Configuration Settings

Configure these options in your VS Code settings.json:

{
  // Ollama Settings
  "aiCodeContextCollector.ollama.enabled": true,
  "aiCodeContextCollector.ollama.endpoint": "http://localhost:11434",
  "aiCodeContextCollector.ollama.model": "qwen2.5-coder:1.5b",
  "aiCodeContextCollector.ollama.timeoutMs": 30000,

  // Graph and Context Budgets
  "aiCodeContextCollector.maxRelationshipDepth": 3,
  "aiCodeContextCollector.maxCandidateFiles": 25,
  "aiCodeContextCollector.maxFiles": 30,
  "aiCodeContextCollector.maxContextSizeKB": 500,
  "aiCodeContextCollector.maxFileSizeKB": 500,

  // Formatting
  "aiCodeContextCollector.includeProjectStructure": true,

  // Exclude Patterns
  "aiCodeContextCollector.excludePatterns": [
    "node_modules",
    ".git",
    "dist",
    "build",
    "coverage",
    ".cache",
    ".angular",
    ".next",
    ".vscode",
    ".idea",
    "vendor",
    "target",
    "bin",
    "obj",
    ".env",
    ".env.*",
    "*.pem",
    "*.key",
    "*.p12",
    "*.pfx",
    "package-lock.json",
    "yarn.lock",
    "pnpm-lock.yaml"
  ]
}

📖 Example Workflow

1. Open your full-stack repository (e.g., Angular frontend + Express backend).
2. Open the Command Palette (Ctrl+Shift+P / Cmd+Shift+P) and run:

   AI Code Context Collector: Create Task Context
   

3. Enter your task:

   Add subscription expiry notification to the customer management page.
   

4. Review the candidate files in the QuickPick dialog, toggle selections if desired, and press Enter.
5. The extension creates and opens ai-task-context.md.
6. Upload or paste ai-task-context.md into your web AI assistant (ChatGPT, Claude, Gemini, Copilot).

📄 License

MIT License. See LICENSE for details.