Description
vibe-pilot is a dedicated context optimizer engineered for AI development workflows, specifically targeting the reduction of token usage in AI prompts. It addresses the common issue of "context tax" by intelligently pruning large codebases, such as those containing lockfiles, generated API clients, vendor trees, and dead boilerplate, into a concise and relevant `.context` file. This process ensures that AI assistants like Cursor, Claude Code, and Windsurf receive only the most pertinent information, drastically cutting token costs and improving the efficiency and accuracy of AI interactions.
Read More
The tool operates locally as a binary with zero network code and no dependencies, ensuring privacy and security as no code is ever uploaded. It performs rapid scans, typically within 15ms, and offers multiple pruning profiles, ranging from an 8k skeleton to a full context window, while always excluding sensitive files like lockfiles, generated code, and binaries. Comments are preserved by default but can be stripped, and secrets are never read or included.
Before generating the `.context` file, vibe-pilot provides a detailed "receipt" that outlines the raw token count, projected cost per prompt based on current model pricing, and per-profile projections. It also lists every skipped file with its reason, offering complete transparency into the optimization process. This allows engineers to understand the exact cost savings and the impact of context optimization.
vibe-pilot integrates seamlessly with AI coding agents through its MCP (Multi-Context Protocol) server functionality. This enables assistants to call `vp_context` for major changes or `vp_diff` between prompts, maintaining an optimized context window automatically. The tool ranks files based on relevance factors such as recent activity, task terms, Git recency, import centrality, and entry-point proximity, ensuring that the most critical code is prioritized. It outputs a single markdown file containing the receipt, a scored map of the repository, and the bodies of the selected files, allowing agents to request any file by path.
Beyond just pruning, vibe-pilot offers a robust alternative to `.gitignore` or traditional packers by providing an itemized receipt and keeping all excluded files visible in the map. It never deletes or modifies original files, only writing a new `.context` file. The tool is compatible with macOS, Linux, and Windows, and is licensed under MIT, making it a versatile and secure solution for AI engineers looking to optimize their development costs and improve AI assistant performance.
- CategoryPrompt Engineering Tools
- Pricing ModelFree
- DeploymentLocal Binary
- PlatformsmacOS, Linux, Windows
vibe-pilot's Core Features
Context Optimization for AI Prompts
Codebase Pruning into .context File
Significant Token Cost Reduction (up to 97.4%)
Zero Dependencies and Local Execution
Fast Scans (15ms typical)
Multiple Pruning Profiles (8k skeleton to full window)
Exclusion of Lockfiles, Generated Code, and Binaries
Detailed Cost Receipt and Token Projections
MCP Native Integration for AI Agents (Claude Code, Cursor)
File Ranking based on Relevance (Git recency, imports, task terms)
Secrets Never Read or Included
Non-destructive Operation (writes only one new file)
Comment Stripping (optional, full-line only)
Cross-Platform Compatibility (macOS, Linux, Windows)
Open-Source (MIT License)
How to use vibe-pilot?
Install vibe-pilot: Download and compile the local binary, which has zero dependencies and compiles quickly.
Analyze your codebase: Run vibe-pilot on your repository to generate a detailed receipt showing raw tokens, projected costs, and potential savings.
Review the receipt: Examine the output to understand what context costs and which files are being cut, along with their reasons.
Generate a .context file: Use vibe-pilot to prune your codebase into a hyper-focused .context file based on your chosen profile.
Feed to AI assistant: Provide the generated .context file to your AI coding assistant (e.g., Cursor, Claude Code, Windsurf).
Integrate with MCP: For continuous optimization, configure your AI agent to call vibe-pilot commands like `vp_context` or `vp_diff` automatically.
vibe-pilot's Use Cases
- Optimizing AI Prompts
- Improving AI Assistant Context
- Cost Management for AI Development
- Secure Code Analysis
- Automated Context Pruning
- Codebase Auditing
